US20240006070A1 - Medical event prediction using a personalized dual-channel combiner network - Google Patents
Medical event prediction using a personalized dual-channel combiner network Download PDFInfo
- Publication number
- US20240006070A1 US20240006070A1 US18/370,062 US202318370062A US2024006070A1 US 20240006070 A1 US20240006070 A1 US 20240006070A1 US 202318370062 A US202318370062 A US 202318370062A US 2024006070 A1 US2024006070 A1 US 2024006070A1
- Authority
- US
- United States
- Prior art keywords
- dccn
- static
- patient
- temporal
- model parameters
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000011282 treatment Methods 0.000 claims abstract description 94
- 238000012549 training Methods 0.000 claims abstract description 72
- 238000000034 method Methods 0.000 claims abstract description 44
- 238000013528 artificial neural network Methods 0.000 claims abstract description 20
- 238000002360 preparation method Methods 0.000 claims abstract description 10
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 7
- 230000002123 temporal effect Effects 0.000 claims description 52
- 230000003068 static effect Effects 0.000 claims description 49
- 238000012545 processing Methods 0.000 claims description 45
- 238000003860 storage Methods 0.000 claims description 42
- 230000006870 function Effects 0.000 claims description 8
- 230000006403 short-term memory Effects 0.000 claims description 5
- 239000013598 vector Substances 0.000 claims description 5
- PRQROPMIIGLWRP-UHFFFAOYSA-N N-formyl-methionyl-leucyl-phenylalanin Chemical compound CSCCC(NC=O)C(=O)NC(CC(C)C)C(=O)NC(C(O)=O)CC1=CC=CC=C1 PRQROPMIIGLWRP-UHFFFAOYSA-N 0.000 claims description 4
- 230000000977 initiatory effect Effects 0.000 claims 3
- 238000000502 dialysis Methods 0.000 description 58
- 238000005259 measurement Methods 0.000 description 34
- 201000010099 disease Diseases 0.000 description 29
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 29
- 238000010586 diagram Methods 0.000 description 22
- 238000007781 pre-processing Methods 0.000 description 18
- 230000015654 memory Effects 0.000 description 14
- 230000008569 process Effects 0.000 description 8
- 230000036772 blood pressure Effects 0.000 description 7
- 238000009534 blood test Methods 0.000 description 6
- 238000004458 analytical method Methods 0.000 description 5
- 238000013480 data collection Methods 0.000 description 5
- 238000012360 testing method Methods 0.000 description 5
- 238000013473 artificial intelligence Methods 0.000 description 4
- 230000008901 benefit Effects 0.000 description 4
- 238000012959 renal replacement therapy Methods 0.000 description 4
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 4
- 238000003491 array Methods 0.000 description 3
- 238000004891 communication Methods 0.000 description 3
- 238000004590 computer program Methods 0.000 description 3
- 230000036541 health Effects 0.000 description 3
- 238000010606 normalization Methods 0.000 description 3
- 230000003287 optical effect Effects 0.000 description 3
- 102000009027 Albumins Human genes 0.000 description 2
- 108010088751 Albumins Proteins 0.000 description 2
- WQZGKKKJIJFFOK-GASJEMHNSA-N Glucose Natural products OC[C@H]1OC(O)[C@H](O)[C@@H](O)[C@@H]1O WQZGKKKJIJFFOK-GASJEMHNSA-N 0.000 description 2
- 208000001953 Hypotension Diseases 0.000 description 2
- 208000007502 anemia Diseases 0.000 description 2
- 239000008280 blood Substances 0.000 description 2
- 210000004369 blood Anatomy 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 230000001627 detrimental effect Effects 0.000 description 2
- 238000009826 distribution Methods 0.000 description 2
- 239000003814 drug Substances 0.000 description 2
- 230000002526 effect on cardiovascular system Effects 0.000 description 2
- 239000008103 glucose Substances 0.000 description 2
- 238000001631 haemodialysis Methods 0.000 description 2
- 230000000322 hemodialysis Effects 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 229910052500 inorganic mineral Inorganic materials 0.000 description 2
- 210000003734 kidney Anatomy 0.000 description 2
- 208000018944 leg cramp Diseases 0.000 description 2
- 239000007788 liquid Substances 0.000 description 2
- 208000012866 low blood pressure Diseases 0.000 description 2
- 230000004060 metabolic process Effects 0.000 description 2
- 239000011707 mineral Substances 0.000 description 2
- 238000012544 monitoring process Methods 0.000 description 2
- 239000004065 semiconductor Substances 0.000 description 2
- 239000007787 solid Substances 0.000 description 2
- 230000032258 transport Effects 0.000 description 2
- 230000006978 adaptation Effects 0.000 description 1
- 230000002411 adverse Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000004140 cleaning Methods 0.000 description 1
- 230000003247 decreasing effect Effects 0.000 description 1
- 230000009977 dual effect Effects 0.000 description 1
- 230000008821 health effect Effects 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 208000017169 kidney disease Diseases 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000000306 recurrent effect Effects 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
- 230000009897 systematic effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H20/00—ICT specially adapted for therapies or health-improving plans, e.g. for handling prescriptions, for steering therapy or for monitoring patient compliance
- G16H20/40—ICT specially adapted for therapies or health-improving plans, e.g. for handling prescriptions, for steering therapy or for monitoring patient compliance relating to mechanical, radiation or invasive therapies, e.g. surgery, laser therapy, dialysis or acupuncture
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
- G16H50/20—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for computer-aided diagnosis, e.g. based on medical expert systems
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
- G06N3/0442—Recurrent networks, e.g. Hopfield networks characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/047—Probabilistic or stochastic networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/096—Transfer learning
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H10/00—ICT specially adapted for the handling or processing of patient-related medical or healthcare data
- G16H10/60—ICT specially adapted for the handling or processing of patient-related medical or healthcare data for patient-specific data, e.g. for electronic patient records
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H40/00—ICT specially adapted for the management or administration of healthcare resources or facilities; ICT specially adapted for the management or operation of medical equipment or devices
- G16H40/60—ICT specially adapted for the management or administration of healthcare resources or facilities; ICT specially adapted for the management or operation of medical equipment or devices for the operation of medical equipment or devices
- G16H40/67—ICT specially adapted for the management or administration of healthcare resources or facilities; ICT specially adapted for the management or operation of medical equipment or devices for the operation of medical equipment or devices for remote operation
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
- G16H50/30—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for calculating health indices; for individual health risk assessment
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
- G16H50/70—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for mining of medical data, e.g. analysing previous cases of other patients
Definitions
- the present invention relates generally to predicting an occurrence of a medical event for a patient, and more particularly, to predicting an occurrence of particular medical events for a patient before, during, and after a medical treatment using a trained neural network.
- a computer implemented method for predicting an occurrence of a medical event for a patient using a trained neural network by preprocessing received historical patient data for a plurality of patients to generate a plurality of normalized training samples.
- Normalized training samples are sent to a personalized deep convolutional neural network (P-DCCN) for model pretraining and updating of model parameters using the P-DCCN, the pretrained model is stored in a remote server for utilization for personalization by a local machine during a preparation time period for a medical treatment, and a normalized finetuning set is generated as output from the P-DCCN by processing input personal data for the patient from the local machine.
- P-DCCN personalized deep convolutional neural network
- Model parameters of the P-DCCN are iteratively finetuned by performing a plurality of training iterations using the generated normalized finetuning set.
- a personalized prediction score for future medical events for the patient is generated using the P-DCCN, and an operation of a medical treatment device is controlled responsive to the personalized prediction score for future medical events.
- P-DCCN personalized deep convolutional neural network
- Model parameters of the P-DCCN are iteratively finetuned by performing a plurality of training iterations using the generated normalized finetuning set.
- a personalized prediction score for future medical events for the patient is generated using the P-DCCN, and an operation of a medical treatment device is controlled responsive to the personalized prediction score for future medical events.
- a non-transitory computer-readable storage medium including a computer-readable program for predicting an occurrence of a medical event for a patient using a trained neural network by preprocessing received historical patient data for a plurality of patients to generate a plurality of normalized training samples.
- Normalized training samples are sent to a personalized deep convolutional neural network (P-DCCN) for model pretraining and updating of model parameters using the P-DCCN, the pretrained model is stored in a remote server for utilization for personalization by a local machine during a preparation time period for a medical treatment, and a normalized finetuning set is generated as output from the P-DCCN by processing input personal data for the patient from the local machine.
- P-DCCN personalized deep convolutional neural network
- Model parameters of the P-DCCN are iteratively finetuned by performing a plurality of training iterations using the generated normalized finetuning set.
- a personalized prediction score for future medical events for the patient is generated using the P-DCCN, and an operation of a medical treatment device is controlled responsive to the personalized prediction score for future medical events.
- FIG. 1 is a block/flow diagram illustrating an exemplary processing system to which the present principles may be applied, in accordance with the present principles
- FIG. 2 shows a diagram of a system/method for predicting disease treatment events using a personalized dual-channel combiner network (P-DCCN), in accordance with an embodiment of the present principles
- FIG. 3 shows a block/flow diagram showing exemplary time periods for prediction of disease treatment events, disease treatment preparation and disease treatment events, in accordance with an embodiment of the present principles
- FIG. 4 shows a diagram of a system/method for predicting disease treatment events using a personalized dual-channel combiner network (P-DCCN), in accordance with an embodiment of the present principles
- FIG. 5 A shows a diagram of a system/method for predicting disease treatment events using a personalized dual-channel combiner network (P-DCCN), in accordance with an embodiment of the present principles
- FIG. 5 B shows a diagram of a system/method for predicting disease treatment events using a personalized dual-channel combiner network (P-DCCN) computing component, in accordance with an embodiment of the present principles
- FIG. 6 shows a diagram of a method for predicting disease treatment events using a personalized dual-channel combiner network (P-DCCN) preprocessing component, in accordance with an embodiment of the present principles
- FIG. 7 shows a high-level block/flow diagram of a system/method for predicting disease treatment events using a personalized dual-channel combiner network (P-DCCN), in accordance with an embodiment of the present principles
- FIG. 8 shows a block/flow diagram of a system/method for predicting disease treatment events using a personalized dual-channel combiner network (P-DCCN), in accordance with an embodiment of the present principles
- FIG. 9 shows a block/flow diagram of a method for predicting disease treatment events using a personalized dual-channel combiner network (P-DCCN), in accordance with an embodiment of the present principles.
- FIG. 10 shows a high-level diagram of a system for conducting medical treatment on a patient, predicting disease treatment events, and monitoring and collecting data from a patient during treatment using a personalized dual-channel combiner network (P-DCCN), in accordance with an embodiment of the present principles.
- P-DCCN personalized dual-channel combiner network
- P-DCCN personalized dual-channel combiner network
- a system and method for predicting an occurrence of particular medical events for a patient before, during, and after a medical treatment (e.g., dialysis treatment) using a trained neural network is provided in accordance with embodiments of the present invention.
- Hemodialysis is a process of purifying the blood of a patient whose kidneys are not working normally, and is one of the important renal replacement therapies (RRT).
- RRT renal replacement therapies
- dialysis patients are at high risk of cardiovascular and other diseases, and thus require intensive management on blood pressure, anemia, mineral metabolism, etc. Otherwise, patients may encounter critical events, such as low blood pressure, leg cramp, and even mortality, during dialysis and/or other medical treatments. Therefore, medical staff decides how to proceed with dialysis from various viewpoints based on patient risks, potential treatment events, and variable clinical factors related to dialysis events.
- the present invention leverages AI systems using a dual-channel combiner network (DCCN) for making prognostic prediction scores during the pre-dialysis period on the incidence of events in future dialysis, which can largely facilitate the decision-making processes of medical staffs, and hence reduce the risk of harmful and/or undesired medical events.
- DCCN dual-channel combiner network
- the present invention harnesses the potential of the management data of dialysis patients by training and utilizing a neural network (e.g., Deep Neural Network (DNN), DCCN, etc.), and thus providing automatic, high-quality, and particularly, personalized prognostic prediction scores on the incidence of events during dialysis, through a new pretraining and finetuning strategy.
- a neural network e.g., Deep Neural Network (DNN), DCCN, etc.
- DNN Deep Neural Network
- DCCN Deep Neural Network
- the present invention is a P-DCCN system/method which provides a systematic and data driven solution to medical event (e.g., dialysis event) prediction not known in the art.
- the present invention is a neural network based intelligent computing system that does not require human efforts or feature engineering.
- the present P-DCCN system/method can include a dual-channel component for integrating static features, low frequent temporal features, and comparatively high frequent temporal features for joint representation learning and the prediction of dialysis events during treatment.
- the present invention utilizes a pre-training and finetuning strategy that addresses and alleviates the challenges of insufficient training data, and the distribution discrepancy of patients data, and thus delivers much higher efficiency of processing and accuracy over conventional models without personalization. This property makes P-DCC N remarkably different from other models with conventional neural network training strategies.
- a pretraining component of the P-DCCN system can be utilized on historical records of a comparatively small amount of a patient's overall medical record data, and can generate a pretrained model that can be stored on server or cloud platform for use for future new patients predictive analysis.
- the finetuning component of the P-DCCN system can send the pre-trained model to local devices where new patients' records are stored for finetuning. This component only uses a comparatively small amount of new records, similarly to the pretraining component described above. With such a small amount of data for personalization, the model can achieve significant improvement of accuracy and decreased processor requirements as compared to conventional, non-personalized systems and methods.
- a dialysis recording data processing component of a DCCN system transforms the historical records of each patient into static profile features and time series features of different frequencies, which can be input to DCCN computing component for further training and/or processing.
- the deep neural network design of the P-DCCN computing component improves prediction accuracy and greatly reduces required human efforts on feature engineering, in accordance with aspects of the present invention.
- the dual-channel design of the P-DCCN computing component can include a multi layer perceptron (MLP) and a long short-term memory (LSTM) recurrent neural network, which can integrate both static features and temporal features of different frequencies for joint event prediction, in accordance with aspects of the present invention.
- MLP multi layer perceptron
- LSTM long short-term memory
- Embodiments described herein may be entirely hardware, entirely software or including both hardware and software elements.
- the present invention is implemented in software, which includes but is not limited to firmware, resident software, microcode, etc.
- Embodiments may include a computer program product accessible from a computer-usable or computer-readable medium providing program code for use by or in connection with a computer or any instruction execution system.
- a computer-usable or computer readable medium may include any apparatus that stores, communicates, propagates, or transports the program for use by or in connection with the instruction execution system, apparatus, or device.
- the medium can be magnetic, optical, electronic, electromagnetic, infrared, or semiconductor system (or apparatus or device) or a propagation medium.
- the medium may include a computer-readable storage medium such as a semiconductor or solid state memory, magnetic tape, a removable computer diskette, a random access memory (RAM), a read-only memory (ROM), a rigid magnetic disk and an optical disk, etc.
- Each computer program may be tangibly stored in a machine-readable storage media or device (e.g., program memory or magnetic disk) readable by a general or special purpose programmable computer, for configuring and controlling operation of a computer when the storage media or device is read by the computer to perform the procedures described herein.
- the inventive system may also be considered to be embodied in a computer-readable storage medium, configured with a computer program, where the storage medium so configured causes a computer to operate in a specific and predefined manner to perform the functions described herein.
- a data processing system suitable for storing and/or executing program code may include at least one processor coupled directly or indirectly to memory elements through a system bus.
- the memory elements can include local memory employed during actual execution of the program code, bulk storage, and cache memories which provide temporary storage of at least some program code to reduce the number of times code is retrieved from bulk storage during execution.
- I/O devices including but not limited to keyboards, displays, pointing devices, etc. may be coupled to the system either directly or through intervening I/O controllers.
- Network adapters may also be coupled to the system to enable the data processing system to become coupled to other data processing systems or remote printers or storage devices through intervening private or public networks.
- Modems, cable modem and Ethernet cards are just a few of the currently available types of network adapters.
- the term “hardware processor subsystem” or “hardware processor” can refer to a processor, memory, software or combinations thereof that cooperate to perform one or more specific tasks.
- the hardware processor subsystem can include one or more data processing elements (e.g., logic circuits, processing circuits, instruction execution devices, etc.).
- the one or more data processing elements can be included in a central processing unit, a graphics processing unit, and/or a separate processor- or computing element-based controller (e.g., logic gates, etc.).
- the hardware processor subsystem can include one or more on-board memories (e.g., caches, dedicated memory arrays, read only memory, etc.).
- the hardware processor subsystem can include one or more memories that can be on or off board or that can be dedicated for use by the hardware processor subsystem (e.g., ROM, RAM, basic input/output system (BIOS), etc.).
- the hardware processor subsystem can include and execute one or more software elements.
- the one or more software elements can include an operating system and/or one or more applications and/or specific code to achieve a specified result.
- the hardware processor subsystem can include dedicated, specialized circuitry that performs one or more electronic processing functions to achieve a specified result.
- Such circuitry can include one or more application-specific integrated circuits (AICs), field-programmable gate arrays (FPGAs), and/or programmable logic arrays (PLAs).
- AICs application-specific integrated circuits
- FPGAs field-programmable gate arrays
- PDAs programmable logic arrays
- the processing system 100 includes at least one processor (CPU) 104 operatively coupled to other components via a system bus 102 .
- a cache 106 operatively coupled to the system bus 102 .
- ROM Read Only Memory
- RAM Random Access Memory
- I/O input/output
- sound adapter 130 operatively coupled to the system bus 102 .
- network adapter 140 operatively coupled to the system bus 102 .
- user interface adapter 150 operatively coupled to the system bus 102 .
- a first storage device 122 and a second storage device 124 are operatively coupled to system bus 102 by the I/O adapter 120 .
- the storage devices 122 and 124 can be any of a disk storage device (e.g., a magnetic or optical disk storage device), a solid state magnetic device, and so forth.
- the storage devices 122 and 124 can be the same type of storage device or different types of storage devices.
- a speaker 132 is operatively coupled to system bus 102 by the sound adapter 130 .
- a transceiver 142 is operatively coupled to system bus 102 by network adapter 140 .
- a display device 162 is operatively coupled to system bus 102 by display adapter 160 .
- a first user input device 152 , a second user input device 154 , and a third user input device 156 are operatively coupled to system bus 102 by user interface adapter 150 .
- the user input devices 152 , 154 , and 156 can be any of a keyboard, a mouse, a keypad, an image capture device, a motion sensing device, a microphone, a device incorporating the functionality of at least two of the preceding devices, and so forth. Of course, other types of input devices can also be used, while maintaining the spirit of the present principles.
- the user input devices 152 , 154 , and 156 can be the same type of user input device or different types of user input devices.
- the user input devices 152 , 154 , and 156 are used to input and output information to and from system 100 .
- processing system 100 may also include other elements (not shown), as readily contemplated by one of skill in the art, as well as omit certain elements.
- various other input devices and/or output devices can be included in processing system 100 , depending upon the particular implementation of the same, as readily understood by one of ordinary skill in the art.
- various types of wireless and/or wired input and/or output devices can be used.
- additional processors, controllers, memories, and so forth, in various configurations can also be utilized as readily appreciated by one of ordinary skill in the art.
- systems 100 , 200 , 400 , 500 , 501 , 700 , 800 , and 1000 are systems for implementing respective embodiments of the present principles.
- Part or all of processing system 100 may be implemented in one or more of the elements of systems 200 , 400 , 500 , 501 , 700 , 800 , and 1000 , according to various embodiments of the present principles.
- processing system 100 may perform at least part of the method described herein including, for example, at least part of methods 300 , 400 , 500 , 501 , 600 , 700 , and 900 of FIGS. 3 , 4 , 5 A, 5 B, 6 , 7 , and 9 , respectively.
- part or all of systems 200 , 400 , 500 , 501 , 700 , 800 , and 1000 may be used to perform at least part of methods 300 , 400 , 500 , 501 , 600 , 700 , and 900 of FIGS. 3 , 4 , 5 A, 5 B, 6 , 7 , and 9 , respectively, according to various embodiments of the present principles.
- FIG. 2 a block diagram showing a system/method 200 for predicting disease treatment events using a personalized dual-channel combiner network (P-DCCN) 218 is illustratively depicted in accordance with an embodiment of the present principles.
- a dual-channel combiner network (DCCN) 202 may be utilized to generate a personalized.
- DCCN (P-DCCN) 218 by training a DCCN 202 using a neural network training component 206 .
- the training component 206 may include a preprocessing component 208 (e.g., for pretraining), a computational component 210 (e.g., for processing static and temporal features), a model storage component 212 (e.g., storage device), a finetuning component 214 , which adapt a globally pretrained model to a particular patient, based on, for example, historical and/or real-time patient measurement devices taken using a patient measurement device 204 and other personal patient data gathered and stored on a local machine 216 , in accordance with aspects of the present invention.
- a preprocessing component 208 e.g., for pretraining
- a computational component 210 e.g., for processing static and temporal features
- a model storage component 212 e.g., storage device
- finetuning component 214 which adapt a globally pretrained model to a particular patient, based on, for example, historical and/or real-time patient measurement devices taken using a patient measurement device 204 and other personal patient data gathered and
- a prediction component 220 can predict future events (e.g., medical treatment events, adverse patient health events, etc.) for a particular patient using the trained P-DCCN, in accordance with aspects of the present invention.
- future events e.g., medical treatment events, adverse patient health events, etc.
- the components of the system 200 may be connected by a bus 201 or may be connected via any suitable communication means (e.g., wireless connection, remote connection across the Internet, wired Ethernet connection, other wired connection, etc.), in accordance with aspects of the present invention.
- FIG. 3 a block/flow diagram showing exemplary time periods for prediction of disease treatment events, disease treatment preparation and disease treatment events 300 is illustratively depicted in accordance with an embodiment of the present principles.
- Medical patients often have a regular routine of receiving treatment for an ailment (e.g., dialysis), generally with treatment at least once per week, depending on the type and/or severity of the ailment.
- ailment e.g., dialysis
- the present invention will be described with reference to dialysis treatment for a patient, although it is to be appreciated that the present principles can be applied to predict medical events for any sort of ailment/disease before, during, and after treatment, in accordance with various embodiments of the present invention.
- Dialysis patients generally have regular routine of dialysis sessions with a frequency of 3 times per week, with each session taking 4 to 5 hours.
- An exemplary one-week schedule 301 for a dialysis patient can include a treatment day 302 , day 2 304 and day 3 306 as non-treatment days, a second treatment day 308 , day 5 310 and day 6 312 as non-treatment days, and a third treatment day 314 .
- the present invention minimizes a risk of detrimental health effects before, during, and/or after a dialysis treatment by predicting the possibility of the incidence of events in a near future dialysis session (e.g., predict events prior to dialysis treatment session) for each patient based on the past recording data utilizing a P-DCCN, in accordance with embodiments of the present invention.
- Historical recording data of dialysis patients mainly constitutes four parts: static profiles of the patients (e.g., age, gender, starting time of dialysis, etc.), dialysis measurement records (with a frequency of 3 times per week (e.g., blood pressure, weight, venous pressure, etc.), blood test measurements (with a frequency of 2 times per month (e.g., albumin, glucose, platelet count, etc.), and cardiothoracic ratio (CTR) (with a frequency of 1 time per month),
- CTR cardiothoracic ratio
- the present invention is an artificial intelligent system, built upon a building block architecture of dual-channel neural networks called dual-channel combiner network (DCCN), which integrates the aforementioned different parts of the data for model training and prognostic score predictions, and will be described in further detail herein below.
- DCCN dual-channel combiner network
- dialysis events e.g., medical events which can occur during dialysis treatment session 309
- the treatment session can be completed in block 311 , and measurements and other patient data can be utilized for further training of the P-DCNN for use at future dialysis treatment sessions (e.g., treatment day 314 ) in accordance with embodiments of the present invention.
- FIG. 4 a block diagram of a system/method 400 for predicting disease treatment events using a personalized dual-channel combiner network (P-DCCN) is illustratively depicted in accordance with an embodiment of the present principles.
- P-DCCN personalized dual-channel combiner network
- the present invention is an artificial intelligence system, built upon a building block architecture of a dual-channel neural network called a dual-channel combiner network (DCCN), which integrates the aforementioned different parts of the data for model training and prognostic score predictions.
- DCCN dual-channel combiner network
- the present invention can generate a personalized DCCN (P-DCCN) for data for all individual patients by utilizing a pretraining and finetuning framework, which is described in further detail herein below, in accordance with aspects of the present invention.
- historical patient records 402 for N patients can be received as input for pretraining 404 to generate a pretrained P-DCCN 406 in accordance with aspects of the present invention.
- the P-DCCN 406 can be trained using historical record data of a plurality of different patients 402 , which can be acquired from a database (e.g., Electronic Health Record (EHR) database), from hospital records for individual patients, etc.
- the historical patient records 402 can include, for example, a patient profile, dialysis measurements, blood test measurements, CTR measurements, etc., which can be utilized as input for training the pretrained P-DCNN 406 in accordance with aspects of the present invention.
- the pretrained P-DCCN 406 can be stored on a server or cloud platform (not shown) for use in further processing.
- the pretrained P-DCCN 406 can be sent to one or more of a plurality of local machines 411 , 413 , 415 by any suitable data transport means (e.g., wireless, wired, remote connection, etc.) for fine tuning data of one or more of a plurality of K new patients (e.g., P N+1 421 , P N+2 423 , . . . , P N+K 425 ) in block 408 .
- K new patients e.g., P N+1 421 , P N+2 423 , . . . , P N+K 425
- the generated, trained personalized P-DCNNs can be used for future predictive analysis for the respective particular patients to predict a probability of medical events occurring before, during, and/or after a medical treatment in blocks 441 , 443 , and 445 , respectively.
- Such personalization of the P-DCNN improves accuracy over conventional models, which do not contemplate such personalization.
- the prediction scores determined in blocks 441 , 443 , and 445 can be output prior to treatment, during treatment, and/or after treatment of a patient, in accordance with aspects of the present invention.
- the P-DCCN 406 can be continuously finetuned in block 408 and personalized for improved accuracy iteratively throughout the treatment of the patient. It is noted that although the present invention was described above with regard to dialysis treatment, it is to be appreciated that the P-DCCN system and method of the present invention can be applied to other medical record data, diseases, and/or medical treatment procedures, in accordance with various embodiments of the present invention.
- FIG. 5 A a high-level diagram of a system/method 500 for predicting disease treatment events using a personalized dual-channel combiner network (P-DCCN) is illustratively depicted in accordance with an embodiment of the present principles.
- P-DCCN personalized dual-channel combiner network
- historical medical records e.g., historic electronic medical records (EMR)
- EMR electronic medical records
- the historical records of dialysis patients can be stored in any suitable form (e.g., cvs, excel, etc.).
- Each patient can have a file that includes medical information on a static profile (e.g., age, gender, starting time of dialysis, etc.), which can be input as static input X in block 506 .
- the file may also include medical information on a temporal profile (e.g., (e.g., dialysis measurements, blood test measurements, event incidences, albumin, glucose, platelet count, etc.), which can be input as temporal input X 1 , X 2 , . . . , X T in block 508 , in accordance with aspects of the present invention.
- a temporal profile e.g., (e.g., dialysis measurements, blood test measurements, event incidences, albumin, glucose, platelet count, etc.)
- Each row can indicate a particular date of a hospital visit by the patient.
- the static input 506 and the temporal input 508 can be sent to a P-DCCN computing component 518 using a static channel 512 and a temporal channel 514 , respectively, for further training and/or processing, in accordance with aspects of the present invention.
- each column can indicate a particular feature, such as some indicator metrics in the dialysis measurements (e.g., blood pressure, weight, venous pressure, etc.). Since different parts have different frequencies, some entries in the form can be blank indicating that feature is not measured at a particular date, in accordance with aspects of the present invention.
- a training label ⁇ can be generated, and sent as input including training loss in block 516 to a P-DCCN computing component 518 , which will be described in further detail herein below with reference to FIG. 5 B .
- the P-DCCN preprocessing component 504 can extract different parts of the data from the files, removes noisy information, and fills in missing values by using mean values of the corresponding features in the historical data and/or by using values from adjacent earlier time steps, in accordance with aspects of the present invention.
- the preprocessing component 504 can set up a time window of width w to segment the time series data, which is described in further detail herein below with reference to FIG. 6 .
- each time window can generate a sample X from time step T ⁇ w to time step T, and can associate it with an event label Y at time step T+1. This generates a sample's focus on the features at the comparatively, closest dates to a future event. Because different parts have different frequencies, all dialysis measurements in the time window can be included, while the blood test measurements on the closest date to the time window can also be included. Then the time window can slide from the beginning of the date to the end of the date in the records to generate multiple samples.
- the preprocessing component 504 can normalize all samples using Gaussian normalization method such that the features of the training samples have mean of 0 and variance of 1, which facilitates the stability of the computing component algorithm in block 518 .
- they can be normalized by using the mean and variance obtained from the training data in block 518 , and the normalized samples can be sent to the model storage component 520 for further model training, testing, and/or storage, in accordance with aspects of the present invention.
- the P-DCCN computing component 518 can include two channels, namely a static channel for processing static and comparatively low frequency temporal features, and a temporal channel for processing comparatively high frequency temporal features, which is described in further detail herein below with reference to FIG. 5 B , in accordance with aspects of the present invention.
- a P-DCCN model storage component 520 can receive a pretrained P-DCCN model as input from the P-DCCN computing component 518 , and the pretrained model can be trained using input historical records 502 of a particular (e.g., threshold) amount of a patient's data provided as the training set. This process can update the parameters of P-DCCN to fit the data in the training set, so that it can extract sufficient knowledge that can be further finetuned and personalized in block 526 using data from a local machine 524 and a P-DCCN personalization component 522 , in accordance with aspects of the present invention.
- the P-DCCN model can be pre-trained using an optimizer with a regression loss function in block 518 as follows:
- ⁇ i is a true indicator of the incidence of an event for the i-th sample in the training data. It is 1 if there is an event, and 0 otherwise.
- ⁇ circumflex over ( ⁇ ) ⁇ i is the predicted score for the i-th sample.
- N is the total number of the training samples.
- ⁇ represents the model parameters.
- ⁇ is a hyperparameter to control the regularization on model parameters for avoid overfitting during the training process.
- the pre-trained P-DCCN (with all parameters updated and fixed) can be sent to a server or a cloud platform for storage in block 520 , so that it can be easily distributed to one or more local machines 524 for further finetuning and personalization in block 526 using a comparatively small amount of records from new patients that are collected by the local machines, in accordance with aspects of the present invention.
- the local machine 524 collects a plurality of different types of records (e.g., static and temporal measurements) for that patient during the time. Although the amount of records collected by the local machine 524 is much smaller than the data size in the pre-training dataset, these records are specific to the particular patient and thus are valuable to adapt the globally pre-trained model to the contexts of the particular patient.
- This personalization process using the P-DCCN personalization component in block 522 via a comparatively small amount of finetuning data leverages the advantages of the few-shot learning, in accordance with aspects of the present invention.
- the pre-trained P-DCCN stored in the model storage component 520 can be sent to a local machine 524 where the finetune dataset can be collected and stored locally.
- the finetune dataset is again preprocessed during the fine tuning in block 526 , similarly to the preprocessing described above with reference to the preprocessing component 504 for generating training samples.
- the pre-trained P-DCCN can be finetuned in block 526 using an optimizer with a regression loss function described above in Section 3 :
- N′ here represents the total number of samples in the finetuning set, which is smaller than N, the number of samples in the pre-training set, where ⁇ i is a true indicator of the incidence of an event for the i-th sample in the training data. It is 1 if there is an event, and 0 otherwise.
- ⁇ circumflex over ( ⁇ ) ⁇ i is the predicted score for the i-th sample.
- N is the total number of the training samples.
- ⁇ represents the model parameters. ⁇ is a hyperparameter to control the regularization on model parameters for avoid overfitting during the training process.
- the personalized model P-DCCN can be used to predict future events for the particular patient by outputting prediction scores for particular events for future time steps 9 based on analysis of the patient's historical records data received as input from the local machine 524 and the input historical EMR data in block 502 .
- Predictions obtained in this manner in block 528 are significantly more accurate than using the pre-trained model directly at least in part because the model is adapted to the particular patient's data so that the distribution discrepancy between the particular patient's data and the data of the pre-training set is alleviated, in accordance with aspects of the present invention.
- FIG. 5 B a diagram 501 of a personalized dual-channel combiner network (P-DCCN) computing component 518 for predicting disease treatment events is illustratively depicted in accordance with an embodiment of the present principles.
- P-DCCN personalized dual-channel combiner network
- the P-DCCN computing component 518 can include two channels, namely a static channel 512 for processing static and comparatively low frequency temporal features, and a temporal channel 514 for processing comparatively high frequency temporal features, in accordance with aspects of the present invention.
- the static channel 512 can receive static features (e.g., liquid temperature, hourly water oval rate, target amount of water removal, dry weight, weight after last dialysis treatment, time before dialysis weight measurement, gain of this time period, etc.) as static input x 506 .
- static features e.g., liquid temperature, hourly water oval rate, target amount of water removal, dry weight, weight after last dialysis treatment, time before dialysis weight measurement, gain of this time period, etc.
- the static features, and comparatively low frequency temporal features can be represented by a vector x s
- the static channel 512 can include a multilayer perceptron 505 (MIT) to encode the information in x s to a compact representation h s by:
- MIT multilayer perceptron 505
- output h s will be a compact representation of the static features (e.g., DNN features 507 ), which can be integrated with the representations from temporal channels for prediction, in accordance with aspects of the present invention.
- a temporal channel 514 can include a plurality of Long Short Term Memory (LSTM) layers 515 A, 515 B, 515 C for processing temporal feature inputs 508 A, 508 B, 508 C, with the temporal feature inputs 508 A, 508 B, 508 C being represented by a sequence of vectors x 1 , x 2 , . . . , x T , respectively.
- the LSTM layers can output a sequence of compact representations 513 , 517 , 512 , 525 h 0 , h 1 , h 2 , . . . , h T respectively, by:
- f LSTM ( ⁇ ) can have multiple layers of LSTM units, which contains trainable model parameters. Also, the LSTM units can be extended to bi-directional LSTM to encode information from both temporal directions in accordance with various embodiments of the present invention,
- h d is the compact representation for all temporal features 535 x 1 , . . . , x T , and is the output of the temporal channel, in accordance with aspects of the present invention.
- the prediction layer 509 can concatenate them and compute the probability of events using an MLP by:
- ⁇ circumflex over ( ⁇ ) ⁇ is a score which indicates the probability of the incidence of a medical event.
- the predicted probability score can be output in block 511 , in accordance with aspects of the present invention.
- FIG. 6 a diagram 600 of a method for predicting disease treatment events using a personalized dual-channel combiner network (P-DCCN) preprocessing component 504 is illustratively depicted in accordance with an embodiment of the present principles.
- P-DCCN personalized dual-channel combiner network
- the diagram 600 illustrates a segmentation process over a time window 604 with reference to dialysis treatment measurements and data.
- Historical dialysis patient measurement data can be input in block 602
- temporal dialysis measurement data can be measured and/or analyzed in blocks 606 and 608 over time during a dialysis treatment.
- Static blood test data 610 and static patient historical healthcare records data 612 can be utilized as input, in addition to static, real-time patient measurement data 614 and/or other dialysis measurement data 616 for prediction of a future medical event 618 , in accordance with aspects of the present invention.
- each time window 604 can generate a sample X from time step T ⁇ w 620 to time step T 622 , and can associate it with an event label Y in block 618 at time step T+1 624 , in accordance with aspects of the present invention.
- This association can generate samples which focus on the features in the closest dates to a future event, and as different parts have different frequencies, all dialysis measurements in the time window 604 can be included, while the blood test measurements 610 on the closest date to the time window can also be included.
- the time window can slide from the beginning of the date to the end of the date (e.g., time period of dialysis treatment, full day, etc.) in the records to generate multiple samples.
- the preprocessing component 504 can normalize all samples using a Gaussian normalization method such that the features of the training samples have mean of 0 and variance of 1, which improves accuracy and stability of the computing algorithm of the P-DCNN computing component 518 , in accordance with aspects of the present invention.
- they can be normalized by using the mean and variance obtained from the training data, and the normalized samples can be sent to the next component e.g., model storage component 520 ) for further model training and testing, in accordance with aspects of the present invention.
- model storage component 520 e.g., model storage component 520
- FIG. 7 a high-level block/flow diagram of a system/method 700 for predicting disease treatment events using a personalized dual-channel combiner network (P-DCCN) is illustratively depicted in accordance with an embodiment of the present principles.
- P-DCCN personalized dual-channel combiner network
- historical patients' pretraining data can be input in block 702 into a pretraining module 704 .
- the pretraining data 702 (e.g., historical recording data of a plurality of patients) can be input to a P-DCCN data preprocessing component 706 and can output normalized samples as the pre-training set, in accordance with aspects of the present invention.
- the normalized samples from block 706 can be sent to a P-DCCN computational component 708 for updating P-DCCN parameters, and can output the pre-trained P-DCCN, and the pretrained P-DCCN from 708 can be sent to a model storage component 710 for future deployment and/or personalization from local machines, in accordance with aspects of the present invention.
- a comparatively small amount of a particular patient's historical medical data (e.g., as compared to the full medial historical record of a patient) can be input from a local machine 712 into a P-DCCN preprocessing component 716 , which can output normalized samples as a finetuning set and be sent to a P-DCCN data collection component 718 , in accordance with aspects of the present invention.
- the pre-trained P-DCCN from the model storage component 710 can be sent to the P-DCCN personalization module 712 , and be utilized by the P-DCCN data collection component 718 for generating personalized prediction scores output in block 720 , in accordance with aspects of the present invention.
- finetuning the model parameters of the pre-trained P-DCCN can be performed by a plurality of training iterations using the finetuning dataset, and the finetuned P-DCCN can be utilized for generating personalized prediction scores ⁇ circumflex over ( ⁇ ) ⁇ in block 720 using the personal data from one or more local machines 714 , in accordance with aspects of the present invention.
- FIG. 8 a block/flow diagram of a system/method 800 for predicting disease treatment events using a personalized dual-channel combiner network (P-DCCN) is illustratively depicted in accordance with an embodiment of the present principles.
- P-DCCN personalized dual-channel combiner network
- a P-DCNN pretraining module 804 can include a P-DCCN preprocessing component 806 , a P-DCNN computational component 814 , and a model storage component 822 .
- the P-DCNN pretraining module 804 can be configured for data cleaning and imputation to improve historical data quality in block 808 , for segmenting recording data and generating time series samples in block 810 , and/or for performing Gaussian normalization of data samples for stability of computation in block 812 , in accordance with aspects of the present invention.
- the P-DCNN computational component 814 can include a dual channel neural network (DCNN) for processing static features and temporal features of different frequencies simultaneously in block 816 , an attention mechanism in temporal channels to learn relative importance of different time steps during integration for performance improvement and interpretation in block 818 , and/or a combination layer to integrate static features and temporal features for computing an event prediction score in block 820 .
- the P-DCCN model storage component 822 can be configured for platform support for running the preprocessing component 806 and the computation component 814 in block 824 , for model pretraining, collection and storage in block 826 , and/or for efficient communication with local machines for sharing pretrained models received as input in block 828 , in accordance with aspects of the present invention.
- a P-DCCN personalization module 830 can include a P-DCCN local data collection component 832 and a P-DCCN finetuning component 840 .
- the local data collection component 832 can be configured for platform support for timely recording and collection of new data from medical treatment (e.g., dialysis) sessions in block 834 , for efficient communication with the model storage component 822 for exchanging of data, including receiving the pretrained model from the model storage component 822 , and to coordinate the running of the finetuning component 840 with the collected data from the local data collection component 832 , in accordance with aspects of the present invention.
- medical treatment e.g., dialysis
- the finetuning component 840 can be configured for collecting the pretrained model and finetuning data in block 842 , for comparatively fast adaptation of the pretrained model to the finetuning data using a few-shot learning strategy in block 844 , for model finetuning, using a regression objective function and/or a gradient optimization algorithm in block 846 , and for generation of personalized prediction scores based on new input data from local machines in block 846 , in accordance with aspects of the present invention.
- FIG. 9 a block/flow diagram 900 of a method for predicting disease treatment events using a personalized dual-channel combiner network (P-DCCN) is illustratively depicted in accordance with an embodiment of the present principles.
- P-DCCN personalized dual-channel combiner network
- historical patients' data can be input in block 902 , and a pretraining set can be generated by preprocessing the historical patients' data in block 904 .
- Normalized samples can be output as a pretraining set to a P-DCCN computational component in block 906 , and the pretrained.
- P-DCCN can be stored in a model storage component in block 908 , in accordance with aspects of the present invention.
- measurements e.g., blood pressure, heart rate, etc.
- a medical treatment e.g., dialysis
- personalized prediction scores can be generated for future medical events for particular patients using the finetuned P-DCCN, in accordance with aspects of the present invention.
- a controller e.g., automatic or manual
- a medical treatment device e.g., dialysis machine
- a plurality of measurement devices e.g., blood pressure monitor, heart rate monitor, etc.
- FIG. 10 a high-level diagram of a system 1000 for conducting medical treatment on a patient, predicting disease treatment events, and monitoring and collecting data from a patient during treatment using a personalized dual-channel combiner network (P-DCCN) is illustratively depicted in accordance with an embodiment of the present principles.
- P-DCCN personalized dual-channel combiner network
- a patient 1001 can be connected to a medical treatment and/or measurement device 1002 (e.g., dialysis machine, an electrocardiogram (EKG) machine, blood pressure monitor, etc.) for receiving a medical treatment by a medical professional 1003 .
- a medical treatment and/or measurement device 1002 e.g., dialysis machine, an electrocardiogram (EKG) machine, blood pressure monitor, etc.
- a P-DCCN system 1006 can be employed for prediction of potential medical events that may occur during the treatment using the medical treatment and/or measurement device 1002 , in accordance with aspects of the present invention.
- the P-DCCN system 106 can be integrated (e.g., built-in) into the medical treatment and/or measurement device 1002 or can be attached via a port (e.g., USB, Ethernet, etc.) to the medical treatment and/or measurement device 1002 such that real-time measurements of the patient 1001 can be taken not only prior to, but also during the treatment for iterative predicting of potential medical events by the P-DCCN system 1006 in real time.
- the medical professional 1003 can utilize a controller (e.g., wired, remote, etc.) to control operation of the medical treatment and/or measurement device 1002 responsive to the event predictions output in real-time by the P-DCCN system in accordance with aspects of the present invention.
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- General Health & Medical Sciences (AREA)
- Public Health (AREA)
- Medical Informatics (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Primary Health Care (AREA)
- Epidemiology (AREA)
- General Physics & Mathematics (AREA)
- Software Systems (AREA)
- Evolutionary Computation (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- Mathematical Physics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- General Engineering & Computer Science (AREA)
- Pathology (AREA)
- Databases & Information Systems (AREA)
- Probability & Statistics with Applications (AREA)
- General Business, Economics & Management (AREA)
- Business, Economics & Management (AREA)
- Urology & Nephrology (AREA)
- Surgery (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Medical Treatment And Welfare Office Work (AREA)
- External Artificial Organs (AREA)
Abstract
Description
- This application is a continuing application of U.S. patent application Ser. No. 17/711,453 filed 1 Apr. 2022, which claims the benefit of Untied States Provisional Patent Application Ser. No. 63/170,660, filed on 5 Apr. 2021, both of which are incorporated by reference in their entireties, incorporated herein by reference in its entirety.
- The present invention relates generally to predicting an occurrence of a medical event for a patient, and more particularly, to predicting an occurrence of particular medical events for a patient before, during, and after a medical treatment using a trained neural network.
- Recently, the tremendous employments of digital systems in hospitals and many medical institutions have brought forth a large volume of healthcare data of patients. The big data are of substantial value, which enables artificial intelligence (AI) to be exploited to support clinical judgement in medicine. As one of the critical themes in modern medicine, the number of patients with kidney diseases has raised social, medical and socioeconomic issues worldwide. Hemodialysis, or simply dialysis, is a process of purifying the blood of a patient whose kidneys are not working normally, and is one of the important renal replacement therapies (RRT). However, dialysis patients are at high risk of cardiovascular and other diseases, and thus require intensive management on blood pressure, anemia, mineral metabolism, etc. Otherwise, patients may encounter critical events, such as low blood pressure, leg cramp, and even mortality, during dialysis and/or other medical treatments.
- A computer implemented method for predicting an occurrence of a medical event for a patient using a trained neural network by preprocessing received historical patient data for a plurality of patients to generate a plurality of normalized training samples. Normalized training samples are sent to a personalized deep convolutional neural network (P-DCCN) for model pretraining and updating of model parameters using the P-DCCN, the pretrained model is stored in a remote server for utilization for personalization by a local machine during a preparation time period for a medical treatment, and a normalized finetuning set is generated as output from the P-DCCN by processing input personal data for the patient from the local machine. Model parameters of the P-DCCN are iteratively finetuned by performing a plurality of training iterations using the generated normalized finetuning set. A personalized prediction score for future medical events for the patient is generated using the P-DCCN, and an operation of a medical treatment device is controlled responsive to the personalized prediction score for future medical events.
- A system for predicting an occurrence of a medical event for a patient using a trained neural network by preprocessing, using a processor operatively coupled to a computer-readable storage medium, received historical patient data for a plurality of patients to generate a plurality of normalized training samples, Normalized training samples are sent to a personalized deep convolutional neural network (P-DCCN) for model pretraining and updating of model parameters using the P-DCCN, the pretrained model is stored in a remote server for utilization for personalization by a local machine during a preparation time period for a medical treatment, and a normalized finetuning set is generated as output from the P-DCCN by processing input personal data for the patient from the local machine. Model parameters of the P-DCCN are iteratively finetuned by performing a plurality of training iterations using the generated normalized finetuning set. A personalized prediction score for future medical events for the patient is generated using the P-DCCN, and an operation of a medical treatment device is controlled responsive to the personalized prediction score for future medical events.
- A non-transitory computer-readable storage medium including a computer-readable program for predicting an occurrence of a medical event for a patient using a trained neural network by preprocessing received historical patient data for a plurality of patients to generate a plurality of normalized training samples. Normalized training samples are sent to a personalized deep convolutional neural network (P-DCCN) for model pretraining and updating of model parameters using the P-DCCN, the pretrained model is stored in a remote server for utilization for personalization by a local machine during a preparation time period for a medical treatment, and a normalized finetuning set is generated as output from the P-DCCN by processing input personal data for the patient from the local machine. Model parameters of the P-DCCN are iteratively finetuned by performing a plurality of training iterations using the generated normalized finetuning set. A personalized prediction score for future medical events for the patient is generated using the P-DCCN, and an operation of a medical treatment device is controlled responsive to the personalized prediction score for future medical events.
- These and other advantages of the invention will be apparent to those of ordinary skill in the art by reference to the following detailed description and the accompanying drawings.
- The disclosure will provide details in the following description of preferred embodiments with reference to the following figures wherein:
-
FIG. 1 is a block/flow diagram illustrating an exemplary processing system to which the present principles may be applied, in accordance with the present principles; -
FIG. 2 shows a diagram of a system/method for predicting disease treatment events using a personalized dual-channel combiner network (P-DCCN), in accordance with an embodiment of the present principles; -
FIG. 3 shows a block/flow diagram showing exemplary time periods for prediction of disease treatment events, disease treatment preparation and disease treatment events, in accordance with an embodiment of the present principles; -
FIG. 4 shows a diagram of a system/method for predicting disease treatment events using a personalized dual-channel combiner network (P-DCCN), in accordance with an embodiment of the present principles; -
FIG. 5A shows a diagram of a system/method for predicting disease treatment events using a personalized dual-channel combiner network (P-DCCN), in accordance with an embodiment of the present principles; -
FIG. 5B shows a diagram of a system/method for predicting disease treatment events using a personalized dual-channel combiner network (P-DCCN) computing component, in accordance with an embodiment of the present principles; -
FIG. 6 shows a diagram of a method for predicting disease treatment events using a personalized dual-channel combiner network (P-DCCN) preprocessing component, in accordance with an embodiment of the present principles; -
FIG. 7 shows a high-level block/flow diagram of a system/method for predicting disease treatment events using a personalized dual-channel combiner network (P-DCCN), in accordance with an embodiment of the present principles; -
FIG. 8 shows a block/flow diagram of a system/method for predicting disease treatment events using a personalized dual-channel combiner network (P-DCCN), in accordance with an embodiment of the present principles; -
FIG. 9 shows a block/flow diagram of a method for predicting disease treatment events using a personalized dual-channel combiner network (P-DCCN), in accordance with an embodiment of the present principles; and -
FIG. 10 shows a high-level diagram of a system for conducting medical treatment on a patient, predicting disease treatment events, and monitoring and collecting data from a patient during treatment using a personalized dual-channel combiner network (P-DCCN), in accordance with an embodiment of the present principles. - In accordance with various embodiments of the present principles, systems and methods are provided for predicting disease treatment events using a personalized dual-channel combiner network (P-DCCN).
- In a particularly useful embodiment, a system and method for predicting an occurrence of particular medical events for a patient before, during, and after a medical treatment (e.g., dialysis treatment) using a trained neural network is provided in accordance with embodiments of the present invention.
- As noted above, Hemodialysis, or simply dialysis, is a process of purifying the blood of a patient whose kidneys are not working normally, and is one of the important renal replacement therapies (RRT). However, dialysis patients are at high risk of cardiovascular and other diseases, and thus require intensive management on blood pressure, anemia, mineral metabolism, etc. Otherwise, patients may encounter critical events, such as low blood pressure, leg cramp, and even mortality, during dialysis and/or other medical treatments. Therefore, medical staff decides how to proceed with dialysis from various viewpoints based on patient risks, potential treatment events, and variable clinical factors related to dialysis events. Given the availability of big medical data, the present invention leverages AI systems using a dual-channel combiner network (DCCN) for making prognostic prediction scores during the pre-dialysis period on the incidence of events in future dialysis, which can largely facilitate the decision-making processes of medical staffs, and hence reduce the risk of harmful and/or undesired medical events.
- However, two key challenges prevent conventional AI systems to be successfully applied for precise analysis of medical data of patients: (1) due to the privacy of data, usually it is difficult to obtain a large amount of patients' data from hospitals that are sufficient for training an accurate model; and (2) due to the high variety of the population among patients, it is difficult for a single pre-trained model to be accurate for every new patient, who are generally different in their age, gender, genetics, health conditions, etc., from the patients data in the training set. As such, a conventional single pre-trained model that is trained a limited training dataset is not generalizable for predictive analysis on data of new patients.
- In accordance with various embodiments, the present invention harnesses the potential of the management data of dialysis patients by training and utilizing a neural network (e.g., Deep Neural Network (DNN), DCCN, etc.), and thus providing automatic, high-quality, and particularly, personalized prognostic prediction scores on the incidence of events during dialysis, through a new pretraining and finetuning strategy. By personalizing a pre-trained model for every patient using a small amount of finetuning data, the model are able to alleviate the aforementioned two challenges and be generalized well to testing dataset and be utilized on patients undergoing medical procedures (e.g., dialysis) to minimize risks and maximize benefits of the procedures for particular patients.
- The present invention is a P-DCCN system/method which provides a systematic and data driven solution to medical event (e.g., dialysis event) prediction not known in the art. The present invention is a neural network based intelligent computing system that does not require human efforts or feature engineering. The present P-DCCN system/method can include a dual-channel component for integrating static features, low frequent temporal features, and comparatively high frequent temporal features for joint representation learning and the prediction of dialysis events during treatment. In various embodiments, the present invention utilizes a pre-training and finetuning strategy that addresses and alleviates the challenges of insufficient training data, and the distribution discrepancy of patients data, and thus delivers much higher efficiency of processing and accuracy over conventional models without personalization. This property makes P-DCC N remarkably different from other models with conventional neural network training strategies.
- In some embodiments, a pretraining component of the P-DCCN system can be utilized on historical records of a comparatively small amount of a patient's overall medical record data, and can generate a pretrained model that can be stored on server or cloud platform for use for future new patients predictive analysis. The finetuning component of the P-DCCN system can send the pre-trained model to local devices where new patients' records are stored for finetuning. This component only uses a comparatively small amount of new records, similarly to the pretraining component described above. With such a small amount of data for personalization, the model can achieve significant improvement of accuracy and decreased processor requirements as compared to conventional, non-personalized systems and methods.
- In some embodiments, a dialysis recording data processing component of a DCCN system transforms the historical records of each patient into static profile features and time series features of different frequencies, which can be input to DCCN computing component for further training and/or processing. The deep neural network design of the P-DCCN computing component improves prediction accuracy and greatly reduces required human efforts on feature engineering, in accordance with aspects of the present invention. In some embodiments, the dual-channel design of the P-DCCN computing component can include a multi layer perceptron (MLP) and a long short-term memory (LSTM) recurrent neural network, which can integrate both static features and temporal features of different frequencies for joint event prediction, in accordance with aspects of the present invention.
- Embodiments described herein may be entirely hardware, entirely software or including both hardware and software elements. In a preferred embodiment, the present invention is implemented in software, which includes but is not limited to firmware, resident software, microcode, etc.
- Embodiments may include a computer program product accessible from a computer-usable or computer-readable medium providing program code for use by or in connection with a computer or any instruction execution system. A computer-usable or computer readable medium may include any apparatus that stores, communicates, propagates, or transports the program for use by or in connection with the instruction execution system, apparatus, or device. The medium can be magnetic, optical, electronic, electromagnetic, infrared, or semiconductor system (or apparatus or device) or a propagation medium. The medium may include a computer-readable storage medium such as a semiconductor or solid state memory, magnetic tape, a removable computer diskette, a random access memory (RAM), a read-only memory (ROM), a rigid magnetic disk and an optical disk, etc.
- Each computer program may be tangibly stored in a machine-readable storage media or device (e.g., program memory or magnetic disk) readable by a general or special purpose programmable computer, for configuring and controlling operation of a computer when the storage media or device is read by the computer to perform the procedures described herein. The inventive system may also be considered to be embodied in a computer-readable storage medium, configured with a computer program, where the storage medium so configured causes a computer to operate in a specific and predefined manner to perform the functions described herein.
- A data processing system suitable for storing and/or executing program code may include at least one processor coupled directly or indirectly to memory elements through a system bus. The memory elements can include local memory employed during actual execution of the program code, bulk storage, and cache memories which provide temporary storage of at least some program code to reduce the number of times code is retrieved from bulk storage during execution. Input/output or I/O devices (including but not limited to keyboards, displays, pointing devices, etc.) may be coupled to the system either directly or through intervening I/O controllers.
- Network adapters may also be coupled to the system to enable the data processing system to become coupled to other data processing systems or remote printers or storage devices through intervening private or public networks. Modems, cable modem and Ethernet cards are just a few of the currently available types of network adapters.
- As employed herein, the term “hardware processor subsystem” or “hardware processor” can refer to a processor, memory, software or combinations thereof that cooperate to perform one or more specific tasks. In useful embodiments, the hardware processor subsystem can include one or more data processing elements (e.g., logic circuits, processing circuits, instruction execution devices, etc.). The one or more data processing elements can be included in a central processing unit, a graphics processing unit, and/or a separate processor- or computing element-based controller (e.g., logic gates, etc.). The hardware processor subsystem can include one or more on-board memories (e.g., caches, dedicated memory arrays, read only memory, etc.). In some embodiments, the hardware processor subsystem can include one or more memories that can be on or off board or that can be dedicated for use by the hardware processor subsystem (e.g., ROM, RAM, basic input/output system (BIOS), etc.).
- In some embodiments, the hardware processor subsystem can include and execute one or more software elements. The one or more software elements can include an operating system and/or one or more applications and/or specific code to achieve a specified result.
- In other embodiments, the hardware processor subsystem can include dedicated, specialized circuitry that performs one or more electronic processing functions to achieve a specified result. Such circuitry can include one or more application-specific integrated circuits (AICs), field-programmable gate arrays (FPGAs), and/or programmable logic arrays (PLAs).
- These and other variations of a hardware processor subsystem are also contemplated in accordance with embodiments of the present invention.
- Referring now to the drawings in which like numerals represent the same or similar elements and initially to
FIG. 1 , anexemplary processing system 100, to which the present principles may be applied, is illustratively depicted in accordance with an embodiment of the present principles. Theprocessing system 100 includes at least one processor (CPU) 104 operatively coupled to other components via asystem bus 102. Acache 106, a Read Only Memory (ROM) 108, a Random Access Memory (RAM) 110, an input/output (I/O)adapter 120, asound adapter 130, anetwork adapter 140, auser interface adapter 150, and adisplay adapter 160, are operatively coupled to thesystem bus 102. - A
first storage device 122 and asecond storage device 124 are operatively coupled tosystem bus 102 by the I/O adapter 120. Thestorage devices storage devices - A
speaker 132 is operatively coupled tosystem bus 102 by thesound adapter 130. Atransceiver 142 is operatively coupled tosystem bus 102 bynetwork adapter 140. Adisplay device 162 is operatively coupled tosystem bus 102 bydisplay adapter 160. - A first
user input device 152, a seconduser input device 154, and a thirduser input device 156 are operatively coupled tosystem bus 102 byuser interface adapter 150. Theuser input devices user input devices user input devices system 100. - Of course, the
processing system 100 may also include other elements (not shown), as readily contemplated by one of skill in the art, as well as omit certain elements. For example, various other input devices and/or output devices can be included inprocessing system 100, depending upon the particular implementation of the same, as readily understood by one of ordinary skill in the art. For example, various types of wireless and/or wired input and/or output devices can be used. Moreover, additional processors, controllers, memories, and so forth, in various configurations can also be utilized as readily appreciated by one of ordinary skill in the art. These and other variations of theprocessing system 100 are readily contemplated by one of ordinary skill in the art given the teachings of the present principles provided herein. - Moreover, it is to be appreciated that
systems FIGS. 1, 2, 4, 5A, 5B, 7, 8, and 10 , respectively, are systems for implementing respective embodiments of the present principles. Part or all ofprocessing system 100 may be implemented in one or more of the elements ofsystems - Further, it is to be appreciated that
processing system 100 may perform at least part of the method described herein including, for example, at least part ofmethods FIGS. 3, 4, 5A, 5B, 6, 7, and 9 , respectively. Similarly, part or all ofsystems methods FIGS. 3, 4, 5A, 5B, 6, 7, and 9 , respectively, according to various embodiments of the present principles. - Referring now to
FIG. 2 , a block diagram showing a system/method 200 for predicting disease treatment events using a personalized dual-channel combiner network (P-DCCN) 218 is illustratively depicted in accordance with an embodiment of the present principles. In one embodiment, a dual-channel combiner network (DCCN) 202 may be utilized to generate a personalized. DCCN (P-DCCN) 218 by training aDCCN 202 using a neuralnetwork training component 206. Thetraining component 206 may include a preprocessing component 208 (e.g., for pretraining), a computational component 210 (e.g., for processing static and temporal features), a model storage component 212 (e.g., storage device), afinetuning component 214, which adapt a globally pretrained model to a particular patient, based on, for example, historical and/or real-time patient measurement devices taken using apatient measurement device 204 and other personal patient data gathered and stored on alocal machine 216, in accordance with aspects of the present invention. - In various embodiments, after the finetuning is completed using the
finetuning component 214, aprediction component 220 can predict future events (e.g., medical treatment events, adverse patient health events, etc.) for a particular patient using the trained P-DCCN, in accordance with aspects of the present invention. It is to be appreciated that the components of thesystem 200 may be connected by abus 201 or may be connected via any suitable communication means (e.g., wireless connection, remote connection across the Internet, wired Ethernet connection, other wired connection, etc.), in accordance with aspects of the present invention. - Referring now to
FIG. 3 , a block/flow diagram showing exemplary time periods for prediction of disease treatment events, disease treatment preparation anddisease treatment events 300 is illustratively depicted in accordance with an embodiment of the present principles. - Medical patients often have a regular routine of receiving treatment for an ailment (e.g., dialysis), generally with treatment at least once per week, depending on the type and/or severity of the ailment. For ease of illustration, the present invention will be described with reference to dialysis treatment for a patient, although it is to be appreciated that the present principles can be applied to predict medical events for any sort of ailment/disease before, during, and after treatment, in accordance with various embodiments of the present invention.
- Dialysis patients generally have regular routine of dialysis sessions with a frequency of 3 times per week, with each session taking 4 to 5 hours. An exemplary one-
week schedule 301 for a dialysis patient can include atreatment day 302,day 2 304 andday 3 306 as non-treatment days, asecond treatment day 308,day 5 310 andday 6 312 as non-treatment days, and athird treatment day 314. As discussed above, various detrimental medical events could occur before, during, and/or after a dialysis treatment, and thus, the present invention minimizes a risk of detrimental health effects before, during, and/or after a dialysis treatment by predicting the possibility of the incidence of events in a near future dialysis session (e.g., predict events prior to dialysis treatment session) for each patient based on the past recording data utilizing a P-DCCN, in accordance with embodiments of the present invention. - Historical recording data of dialysis patients mainly constitutes four parts: static profiles of the patients (e.g., age, gender, starting time of dialysis, etc.), dialysis measurement records (with a frequency of 3 times per week (e.g., blood pressure, weight, venous pressure, etc.), blood test measurements (with a frequency of 2 times per month (e.g., albumin, glucose, platelet count, etc.), and cardiothoracic ratio (CTR) (with a frequency of 1 time per month), The last three parts are dynamic and change over time, so they can be modeled by time series, but with different frequencies, in accordance with aspects of the present invention.
- The present invention is an artificial intelligent system, built upon a building block architecture of dual-channel neural networks called dual-channel combiner network (DCCN), which integrates the aforementioned different parts of the data for model training and prognostic score predictions, and will be described in further detail herein below.
- In accordance with embodiments of the present invention, during a
treatment day treatment 307 prior to conducting thetreatment session 309. The treatment session can be completed inblock 311, and measurements and other patient data can be utilized for further training of the P-DCNN for use at future dialysis treatment sessions (e.g., treatment day 314) in accordance with embodiments of the present invention. - Referring now to
FIG. 4 , a block diagram of a system/method 400 for predicting disease treatment events using a personalized dual-channel combiner network (P-DCCN) is illustratively depicted in accordance with an embodiment of the present principles. - In some embodiments, the present invention is an artificial intelligence system, built upon a building block architecture of a dual-channel neural network called a dual-channel combiner network (DCCN), which integrates the aforementioned different parts of the data for model training and prognostic score predictions. In various embodiments, the present invention can generate a personalized DCCN (P-DCCN) for data for all individual patients by utilizing a pretraining and finetuning framework, which is described in further detail herein below, in accordance with aspects of the present invention.
- In some embodiments,
historical patient records 402 for N patients (e.g.,P 1 401,P 2 403, . . . , PN 405) can be received as input for pretraining 404 to generate a pretrained P-DCCN 406 in accordance with aspects of the present invention. In thepretraining stage 404, the P-DCCN 406 can be trained using historical record data of a plurality ofdifferent patients 402, which can be acquired from a database (e.g., Electronic Health Record (EHR) database), from hospital records for individual patients, etc. Thehistorical patient records 402 can include, for example, a patient profile, dialysis measurements, blood test measurements, CTR measurements, etc., which can be utilized as input for training the pretrained P-DCNN 406 in accordance with aspects of the present invention. - In some embodiments, the pretrained P-
DCCN 406 can be stored on a server or cloud platform (not shown) for use in further processing. The pretrained P-DCCN 406 can be sent to one or more of a plurality oflocal machines P N+1 421,P N+2 423, . . . , PN+K 425) inblock 408. At the finetuningstage 408, once a new patient has accumulated certain amount of medical record data (e.g., a comparatively small amount from a patient's overall medical record), such as several weeks of records, these records can be used to finetune the pre-trained P-DCCN 406 for personalization for particular patients. - In some embodiments, the generated, trained personalized P-DCNNs (e.g.,
P N+1 431,P N+2 433, . . . , PN+K 435) can be used for future predictive analysis for the respective particular patients to predict a probability of medical events occurring before, during, and/or after a medical treatment inblocks blocks DCCN 406 can be continuously finetuned inblock 408 and personalized for improved accuracy iteratively throughout the treatment of the patient. It is noted that although the present invention was described above with regard to dialysis treatment, it is to be appreciated that the P-DCCN system and method of the present invention can be applied to other medical record data, diseases, and/or medical treatment procedures, in accordance with various embodiments of the present invention. - Referring now to
FIG. 5A , a high-level diagram of a system/method 500 for predicting disease treatment events using a personalized dual-channel combiner network (P-DCCN) is illustratively depicted in accordance with an embodiment of the present principles. - In accordance with embodiments of the present invention, historical medical records (e.g., historic electronic medical records (EMR)) can be input as a neural network training set in
block 502 for use by the P-DCNNdata preprocessing component 504. The historical records of dialysis patients can be stored in any suitable form (e.g., cvs, excel, etc.). Each patient can have a file that includes medical information on a static profile (e.g., age, gender, starting time of dialysis, etc.), which can be input as static input X inblock 506. The file may also include medical information on a temporal profile (e.g., (e.g., dialysis measurements, blood test measurements, event incidences, albumin, glucose, platelet count, etc.), which can be input as temporal input X1, X2, . . . , XT inblock 508, in accordance with aspects of the present invention. Each row can indicate a particular date of a hospital visit by the patient. In some embodiments, thestatic input 506 and thetemporal input 508 can be sent to a P-DCCN computing component 518 using astatic channel 512 and atemporal channel 514, respectively, for further training and/or processing, in accordance with aspects of the present invention. - In some embodiments, each column can indicate a particular feature, such as some indicator metrics in the dialysis measurements (e.g., blood pressure, weight, venous pressure, etc.). Since different parts have different frequencies, some entries in the form can be blank indicating that feature is not measured at a particular date, in accordance with aspects of the present invention. In
block 510, a training label γ can be generated, and sent as input including training loss inblock 516 to a P-DCCN computing component 518, which will be described in further detail herein below with reference toFIG. 5B . - In various embodiments, the P-
DCCN preprocessing component 504 can extract different parts of the data from the files, removes noisy information, and fills in missing values by using mean values of the corresponding features in the historical data and/or by using values from adjacent earlier time steps, in accordance with aspects of the present invention. - In some embodiments, the
preprocessing component 504 can set up a time window of width w to segment the time series data, which is described in further detail herein below with reference toFIG. 6 . In an embodiment, each time window can generate a sample X from time step T−w to time step T, and can associate it with an event label Y at timestep T+ 1. This generates a sample's focus on the features at the comparatively, closest dates to a future event. Because different parts have different frequencies, all dialysis measurements in the time window can be included, while the blood test measurements on the closest date to the time window can also be included. Then the time window can slide from the beginning of the date to the end of the date in the records to generate multiple samples. - In an embodiment, after samples are generated, the
preprocessing component 504 can normalize all samples using Gaussian normalization method such that the features of the training samples have mean of 0 and variance of 1, which facilitates the stability of the computing component algorithm inblock 518. For testing samples, they can be normalized by using the mean and variance obtained from the training data inblock 518, and the normalized samples can be sent to themodel storage component 520 for further model training, testing, and/or storage, in accordance with aspects of the present invention. In various embodiment, the P-DCCN computing component 518 can include two channels, namely a static channel for processing static and comparatively low frequency temporal features, and a temporal channel for processing comparatively high frequency temporal features, which is described in further detail herein below with reference toFIG. 5B , in accordance with aspects of the present invention. - In some embodiments, a P-DCCN
model storage component 520 can receive a pretrained P-DCCN model as input from the P-DCCN computing component 518, and the pretrained model can be trained using inputhistorical records 502 of a particular (e.g., threshold) amount of a patient's data provided as the training set. This process can update the parameters of P-DCCN to fit the data in the training set, so that it can extract sufficient knowledge that can be further finetuned and personalized inblock 526 using data from alocal machine 524 and a P-DCCN personalization component 522, in accordance with aspects of the present invention. - In various embodiments, the P-DCCN model can be pre-trained using an optimizer with a regression loss function in
block 518 as follows: -
- where γi is a true indicator of the incidence of an event for the i-th sample in the training data. It is 1 if there is an event, and 0 otherwise. {circumflex over (γ)}i is the predicted score for the i-th sample. N is the total number of the training samples. θ represents the model parameters. λ is a hyperparameter to control the regularization on model parameters for avoid overfitting during the training process.
- After pre-training is done in
block 518, the pre-trained P-DCCN (with all parameters updated and fixed) can be sent to a server or a cloud platform for storage inblock 520, so that it can be easily distributed to one or morelocal machines 524 for further finetuning and personalization inblock 526 using a comparatively small amount of records from new patients that are collected by the local machines, in accordance with aspects of the present invention. - In practice, when a new patient has been attending dialysis treatments for several weeks, the
local machine 524 collects a plurality of different types of records (e.g., static and temporal measurements) for that patient during the time. Although the amount of records collected by thelocal machine 524 is much smaller than the data size in the pre-training dataset, these records are specific to the particular patient and thus are valuable to adapt the globally pre-trained model to the contexts of the particular patient. This personalization process using the P-DCCN personalization component inblock 522 via a comparatively small amount of finetuning data leverages the advantages of the few-shot learning, in accordance with aspects of the present invention. - In some embodiments the pre-trained P-DCCN stored in the
model storage component 520 can be sent to alocal machine 524 where the finetune dataset can be collected and stored locally. The finetune dataset is again preprocessed during the fine tuning inblock 526, similarly to the preprocessing described above with reference to thepreprocessing component 504 for generating training samples. - In some embodiments, the pre-trained P-DCCN can be finetuned in
block 526 using an optimizer with a regression loss function described above in Section 3: -
- where N′ here represents the total number of samples in the finetuning set, which is smaller than N, the number of samples in the pre-training set, where γi is a true indicator of the incidence of an event for the i-th sample in the training data. It is 1 if there is an event, and 0 otherwise. {circumflex over (γ)}i is the predicted score for the i-th sample. N is the total number of the training samples. θ represents the model parameters. λ is a hyperparameter to control the regularization on model parameters for avoid overfitting during the training process.
- In some embodiments, once the finetuning in
block 526 is done, the personalized model P-DCCN can be used to predict future events for the particular patient by outputting prediction scores for particular events for future time steps 9 based on analysis of the patient's historical records data received as input from thelocal machine 524 and the input historical EMR data inblock 502. Predictions obtained in this manner inblock 528 are significantly more accurate than using the pre-trained model directly at least in part because the model is adapted to the particular patient's data so that the distribution discrepancy between the particular patient's data and the data of the pre-training set is alleviated, in accordance with aspects of the present invention. - Referring now to
FIG. 5B , a diagram 501 of a personalized dual-channel combiner network (P-DCCN)computing component 518 for predicting disease treatment events is illustratively depicted in accordance with an embodiment of the present principles. - In various embodiments, the P-
DCCN computing component 518 can include two channels, namely astatic channel 512 for processing static and comparatively low frequency temporal features, and atemporal channel 514 for processing comparatively high frequency temporal features, in accordance with aspects of the present invention. - In some embodiments the
static channel 512 can receive static features (e.g., liquid temperature, hourly water oval rate, target amount of water removal, dry weight, weight after last dialysis treatment, time before dialysis weight measurement, gain of this time period, etc.) as static input x 506. The static features, and comparatively low frequency temporal features, can be represented by a vector xs, and thestatic channel 512 can include a multilayer perceptron 505 (MIT) to encode the information in xs to a compact representation hs by: -
h s =f MLP(x s) - where fMLP(⋅) can be multiple layers of fully connected network with the form Wsxs+bs, with Ws and bs being the model parameters to be trained, in accordance with aspects of the present invention. In some embodiments, output hs will be a compact representation of the static features (e.g., DNN features 507), which can be integrated with the representations from temporal channels for prediction, in accordance with aspects of the present invention.
- In various embodiments, a
temporal channel 514 can include a plurality of Long Short Term Memory (LSTM) layers 515A, 515B, 515C for processingtemporal feature inputs temporal feature inputs compact representations -
h 1 , . . . ,h T =f LSTM(x 1 , . . . ,x T) - where fLSTM (⋅) can have multiple layers of LSTM units, which contains trainable model parameters. Also, the LSTM units can be extended to bi-directional LSTM to encode information from both temporal directions in accordance with various embodiments of the present invention,
-
- In some embodiments, on top of the LSTM layers 515A, 515B, 515C,
compact representations attention layer attention layer blocks
- In some embodiments, on top of the LSTM layers 515A, 515B, 515C,
-
e t =w αtan h(W α h t) for t=1, . . . ,T -
αt=softmax(e t) for t=1, . . . ,T - where Wα and wα are model parameters to learn. After this step, Σt=1 Tαt=1. Then, all compact temporal representations can be combined (e.g., using a Hadamard product in
blocks attention weights -
- where hd is the compact representation for all temporal features 535 x1, . . . , xT, and is the output of the temporal channel, in accordance with aspects of the present invention.
- In various embodiments, after the static and temporal representations hs and hd are obtained from the
static channel 512 andtemporal channel 514, theprediction layer 509 can concatenate them and compute the probability of events using an MLP by: -
{circumflex over (γ)}=f MLP([h s ,h d]) - where {circumflex over (γ)} is a score which indicates the probability of the incidence of a medical event. The predicted probability score can be output in block 511, in accordance with aspects of the present invention.
- Referring now to
FIG. 6 , with continued reference toFIG. 5A , a diagram 600 of a method for predicting disease treatment events using a personalized dual-channel combiner network (P-DCCN) preprocessingcomponent 504 is illustratively depicted in accordance with an embodiment of the present principles. - In accordance with embodiments of the present invention, the diagram 600 illustrates a segmentation process over a
time window 604 with reference to dialysis treatment measurements and data. Historical dialysis patient measurement data can be input inblock 602, and temporal dialysis measurement data can be measured and/or analyzed inblocks blood test data 610 and static patient historicalhealthcare records data 612 can be utilized as input, in addition to static, real-timepatient measurement data 614 and/or otherdialysis measurement data 616 for prediction of a futuremedical event 618, in accordance with aspects of the present invention. - In some embodiments, each
time window 604 can generate a sample X from time step T−w 620 totime step T 622, and can associate it with an event label Y inblock 618 at time step T+1 624, in accordance with aspects of the present invention. This association can generate samples which focus on the features in the closest dates to a future event, and as different parts have different frequencies, all dialysis measurements in thetime window 604 can be included, while theblood test measurements 610 on the closest date to the time window can also be included. The time window can slide from the beginning of the date to the end of the date (e.g., time period of dialysis treatment, full day, etc.) in the records to generate multiple samples. - After samples are generated, the
preprocessing component 504 can normalize all samples using a Gaussian normalization method such that the features of the training samples have mean of 0 and variance of 1, which improves accuracy and stability of the computing algorithm of the P-DCNN computing component 518, in accordance with aspects of the present invention. For testing samples, they can be normalized by using the mean and variance obtained from the training data, and the normalized samples can be sent to the next component e.g., model storage component 520) for further model training and testing, in accordance with aspects of the present invention. In practice, some of the dialysis measurements can be evaluated on the same date for which event is to be predicted. These measurements (e.g., liquid temperature, hourly water removal rate, target amount of water removal, thy weight, weight after last dialysis treatment, time before dialysis weight measurement, gain of this time period, etc.) can be evaluated immediately before the dialysis starts, and thus can be included asstatic features 614 for further processing, in accordance with aspects of the present invention. - Referring now to
FIG. 7 , a high-level block/flow diagram of a system/method 700 for predicting disease treatment events using a personalized dual-channel combiner network (P-DCCN) is illustratively depicted in accordance with an embodiment of the present principles. - In accordance with various embodiments, historical patients' pretraining data can be input in
block 702 into apretraining module 704. The pretraining data 702 (e.g., historical recording data of a plurality of patients) can be input to a P-DCCNdata preprocessing component 706 and can output normalized samples as the pre-training set, in accordance with aspects of the present invention. The normalized samples fromblock 706 can be sent to a P-DCCNcomputational component 708 for updating P-DCCN parameters, and can output the pre-trained P-DCCN, and the pretrained P-DCCN from 708 can be sent to amodel storage component 710 for future deployment and/or personalization from local machines, in accordance with aspects of the present invention. - In an embodiment, using a
personalization module 712, a comparatively small amount of a particular patient's historical medical data (e.g., as compared to the full medial historical record of a patient) can be input from alocal machine 712 into a P-DCCN preprocessing component 716, which can output normalized samples as a finetuning set and be sent to a P-DCCNdata collection component 718, in accordance with aspects of the present invention. The pre-trained P-DCCN from themodel storage component 710 can be sent to the P-DCCN personalization module 712, and be utilized by the P-DCCNdata collection component 718 for generating personalized prediction scores output inblock 720, in accordance with aspects of the present invention. In some embodiments, finetuning the model parameters of the pre-trained P-DCCN can be performed by a plurality of training iterations using the finetuning dataset, and the finetuned P-DCCN can be utilized for generating personalized prediction scores {circumflex over (γ)} inblock 720 using the personal data from one or morelocal machines 714, in accordance with aspects of the present invention. - Referring now to
FIG. 8 , a block/flow diagram of a system/method 800 for predicting disease treatment events using a personalized dual-channel combiner network (P-DCCN) is illustratively depicted in accordance with an embodiment of the present principles. - In accordance with embodiments of the present invention, a system architecture of a P-
DCCN system 802 is provided. A P-DCNN pretraining module 804 can include a P-DCCN preprocessing component 806, a P-DCNNcomputational component 814, and amodel storage component 822. In some embodiments, the P-DCNN pretraining module 804 can be configured for data cleaning and imputation to improve historical data quality inblock 808, for segmenting recording data and generating time series samples inblock 810, and/or for performing Gaussian normalization of data samples for stability of computation inblock 812, in accordance with aspects of the present invention. - In some embodiments, the P-DCNN
computational component 814 can include a dual channel neural network (DCNN) for processing static features and temporal features of different frequencies simultaneously in block 816, an attention mechanism in temporal channels to learn relative importance of different time steps during integration for performance improvement and interpretation inblock 818, and/or a combination layer to integrate static features and temporal features for computing an event prediction score inblock 820. In some embodiments, the P-DCCNmodel storage component 822 can be configured for platform support for running thepreprocessing component 806 and thecomputation component 814 inblock 824, for model pretraining, collection and storage inblock 826, and/or for efficient communication with local machines for sharing pretrained models received as input inblock 828, in accordance with aspects of the present invention. - In accordance with various embodiments, a P-
DCCN personalization module 830 can include a P-DCCN localdata collection component 832 and a P-DCCN finetuning component 840. The localdata collection component 832 can be configured for platform support for timely recording and collection of new data from medical treatment (e.g., dialysis) sessions inblock 834, for efficient communication with themodel storage component 822 for exchanging of data, including receiving the pretrained model from themodel storage component 822, and to coordinate the running of thefinetuning component 840 with the collected data from the localdata collection component 832, in accordance with aspects of the present invention. - In some embodiments, the finetuning
component 840 can be configured for collecting the pretrained model and finetuning data inblock 842, for comparatively fast adaptation of the pretrained model to the finetuning data using a few-shot learning strategy inblock 844, for model finetuning, using a regression objective function and/or a gradient optimization algorithm inblock 846, and for generation of personalized prediction scores based on new input data from local machines inblock 846, in accordance with aspects of the present invention. - Referring now to
FIG. 9 , a block/flow diagram 900 of a method for predicting disease treatment events using a personalized dual-channel combiner network (P-DCCN) is illustratively depicted in accordance with an embodiment of the present principles. - In accordance with various embodiments, historical patients' data can be input in
block 902, and a pretraining set can be generated by preprocessing the historical patients' data inblock 904. Normalized samples can be output as a pretraining set to a P-DCCN computational component inblock 906, and the pretrained. P-DCCN can be stored in a model storage component inblock 908, in accordance with aspects of the present invention. Inblock 910, measurements (e.g., blood pressure, heart rate, etc.) can be taken of a patient before, during, and/or after a medical treatment (e.g., dialysis), and iterative finetuning, and personalization of the P-DCNN can be performed based on the measurements fromblock 910 and other data stored on local machines in block 912. Inblock 914, personalized prediction scores can be generated for future medical events for particular patients using the finetuned P-DCCN, in accordance with aspects of the present invention. Inblock 916, a controller (e.g., automatic or manual) can be utilized to control operation of a medical treatment device (e.g., dialysis machine) and/or a plurality of measurement devices (e.g., blood pressure monitor, heart rate monitor, etc.) responsive to the personalized prediction scores generated inblock 914, in accordance with aspects of the present invention. - Referring now to
FIG. 10 , a high-level diagram of asystem 1000 for conducting medical treatment on a patient, predicting disease treatment events, and monitoring and collecting data from a patient during treatment using a personalized dual-channel combiner network (P-DCCN) is illustratively depicted in accordance with an embodiment of the present principles. - In accordance with various embodiments of the present invention, a
patient 1001 can be connected to a medical treatment and/or measurement device 1002 (e.g., dialysis machine, an electrocardiogram (EKG) machine, blood pressure monitor, etc.) for receiving a medical treatment by a medical professional 1003. Nor to the medical treatment, a P-DCCN system 1006 can be employed for prediction of potential medical events that may occur during the treatment using the medical treatment and/ormeasurement device 1002, in accordance with aspects of the present invention. The P-DCCN system 106 can be integrated (e.g., built-in) into the medical treatment and/ormeasurement device 1002 or can be attached via a port (e.g., USB, Ethernet, etc.) to the medical treatment and/ormeasurement device 1002 such that real-time measurements of thepatient 1001 can be taken not only prior to, but also during the treatment for iterative predicting of potential medical events by the P-DCCN system 1006 in real time. The medical professional 1003 can utilize a controller (e.g., wired, remote, etc.) to control operation of the medical treatment and/ormeasurement device 1002 responsive to the event predictions output in real-time by the P-DCCN system in accordance with aspects of the present invention. - The foregoing is to be understood as being in every respect illustrative and exemplary, but not restrictive, and the scope of the invention disclosed herein is not to be determined from the Detailed Description, but rather from the claims as interpreted according to the full breadth permitted by the patent laws. It is to be understood that the embodiments shown and described herein are only illustrative of the principles of the present invention and that those skilled in the art may implement various modifications without departing from the scope and spirit of the invention. Those skilled in the art could implement various other feature combinations without departing from the scope and spirit of the invention. Having thus described aspects of the invention, with the details and particularity required by the patent laws, what is claimed and desired protected by Letters Patent is set forth in the appended claims.
Claims (18)
h s =f MLP(x s)
h 1 , . . . ,h T =f LSTM(x 1 , . . . ,x T)
{circumflex over (γ)}=f MLP([h s ,h d])
h s =f MLP(x s)
h 1 , . . . ,h T =f LSTM(x 1 , . . . ,x T)
{circumflex over (γ)}=f MLP([h s ,h d])
h s =f MLP(x s)
h 1 , . . . ,h T =f LSTM(x 1 , . . . ,x T)
{circumflex over (γ)}=f MLP([h s ,h d])
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
US18/370,062 US20240006070A1 (en) | 2021-04-05 | 2023-09-19 | Medical event prediction using a personalized dual-channel combiner network |
Applications Claiming Priority (3)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
US202163170660P | 2021-04-05 | 2021-04-05 | |
US17/711,453 US20220319709A1 (en) | 2021-04-05 | 2022-04-01 | Medical event prediction using a personalized dual-channel combiner network |
US18/370,062 US20240006070A1 (en) | 2021-04-05 | 2023-09-19 | Medical event prediction using a personalized dual-channel combiner network |
Related Parent Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
US17/711,453 Continuation US20220319709A1 (en) | 2021-04-05 | 2022-04-01 | Medical event prediction using a personalized dual-channel combiner network |
Publications (1)
Publication Number | Publication Date |
---|---|
US20240006070A1 true US20240006070A1 (en) | 2024-01-04 |
Family
ID=83450043
Family Applications (4)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
US17/711,453 Pending US20220319709A1 (en) | 2021-04-05 | 2022-04-01 | Medical event prediction using a personalized dual-channel combiner network |
US18/370,062 Pending US20240006070A1 (en) | 2021-04-05 | 2023-09-19 | Medical event prediction using a personalized dual-channel combiner network |
US18/370,074 Pending US20240013920A1 (en) | 2021-04-05 | 2023-09-19 | Medical event prediction using a personalized dual-channel combiner network |
US18/370,049 Pending US20240006069A1 (en) | 2021-04-05 | 2023-09-19 | Medical Event Prediction Using a Personalized Dual-Channel Combiner Network |
Family Applications Before (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
US17/711,453 Pending US20220319709A1 (en) | 2021-04-05 | 2022-04-01 | Medical event prediction using a personalized dual-channel combiner network |
Family Applications After (2)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
US18/370,074 Pending US20240013920A1 (en) | 2021-04-05 | 2023-09-19 | Medical event prediction using a personalized dual-channel combiner network |
US18/370,049 Pending US20240006069A1 (en) | 2021-04-05 | 2023-09-19 | Medical Event Prediction Using a Personalized Dual-Channel Combiner Network |
Country Status (4)
Country | Link |
---|---|
US (4) | US20220319709A1 (en) |
JP (1) | JP2024518693A (en) |
DE (1) | DE112022001973T5 (en) |
WO (1) | WO2022216618A1 (en) |
Families Citing this family (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2024089928A1 (en) * | 2022-10-28 | 2024-05-02 | 株式会社レナサイエンス | Method for assisting advance prediction of target dewatering amount in dialysis, assisting device, and assisting program |
Family Cites Families (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20140108035A1 (en) * | 2012-10-11 | 2014-04-17 | Kunter Seref Akbay | System and method to automatically assign resources in a network of healthcare enterprises |
US9730643B2 (en) * | 2013-10-17 | 2017-08-15 | Siemens Healthcare Gmbh | Method and system for anatomical object detection using marginal space deep neural networks |
EP3255573A1 (en) * | 2016-06-10 | 2017-12-13 | Electronics and Telecommunications Research Institute | Clinical decision supporting ensemble system and clinical decison supporting method using the same |
WO2019022779A1 (en) * | 2017-07-28 | 2019-01-31 | Google Llc | System and method for predicting and summarizing medical events from electronic health records |
US11756667B2 (en) * | 2018-05-30 | 2023-09-12 | Siemens Healthcare Gmbh | Decision support system for medical therapy planning |
-
2022
- 2022-04-01 US US17/711,453 patent/US20220319709A1/en active Pending
- 2022-04-04 WO PCT/US2022/023327 patent/WO2022216618A1/en active Application Filing
- 2022-04-04 JP JP2023561291A patent/JP2024518693A/en active Pending
- 2022-04-04 DE DE112022001973.5T patent/DE112022001973T5/en active Pending
-
2023
- 2023-09-19 US US18/370,062 patent/US20240006070A1/en active Pending
- 2023-09-19 US US18/370,074 patent/US20240013920A1/en active Pending
- 2023-09-19 US US18/370,049 patent/US20240006069A1/en active Pending
Also Published As
Publication number | Publication date |
---|---|
US20240006069A1 (en) | 2024-01-04 |
US20240013920A1 (en) | 2024-01-11 |
DE112022001973T5 (en) | 2024-01-18 |
WO2022216618A1 (en) | 2022-10-13 |
JP2024518693A (en) | 2024-05-02 |
US20220319709A1 (en) | 2022-10-06 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Kamalraj et al. | Interpretable filter based convolutional neural network (IF-CNN) for glucose prediction and classification using PD-SS algorithm | |
Ambekar et al. | Disease risk prediction by using convolutional neural network | |
CN109599177B (en) | Method for predicting medical treatment track through deep learning based on medical history | |
CN112289442B (en) | Method and device for predicting disease end point event and electronic equipment | |
Akter et al. | Comprehensive performance assessment of deep learning models in early prediction and risk identification of chronic kidney disease | |
CN109830303A (en) | Clinical data mining analysis and aid decision-making method based on internet integration medical platform | |
Acharjya | A hybrid scheme for heart disease diagnosis using rough set and cuckoo search technique | |
US20240006078A1 (en) | Left ventricular volume and cardiac output estimation using machine learning model | |
US20240006070A1 (en) | Medical event prediction using a personalized dual-channel combiner network | |
Ma et al. | A general framework for diagnosis prediction via incorporating medical code descriptions | |
Zhu et al. | Blood glucose prediction for type 1 diabetes using generative adversarial networks | |
WO2021158796A1 (en) | Forecasting and explaining user health metrics | |
Rodrigues-Jr et al. | LIG-Doctor: Efficient patient trajectory prediction using bidirectional minimal gated-recurrent networks | |
WO2022046734A1 (en) | Robust forecasting system on irregular time series in dialysis medical records | |
Kirubakaran et al. | Echo state learned compositional pattern neural networks for the early diagnosis of cancer on the internet of medical things platform | |
US20220318626A1 (en) | Meta-training framework on dual-channel combiner network system for dialysis event prediction | |
CN117672451A (en) | Medicine recommendation method for type 2 diabetics | |
Portela et al. | Real-time Intelligent decision support in intensive medicine | |
Habibi et al. | Estimation of re-hospitalization risk of diabetic patients based on radial base function (RBF) neural network method combined with colonial competition optimization algorithm | |
Luo et al. | GlucoGuide: an intelligent type-2 diabetes solution using data mining and mobile computing | |
Ismail et al. | Multivariate multi-step deep learning time series approach in forecasting Parkinson's disease future severity progression | |
Usha et al. | Revived ant colony optimization-based adaboost algorithm for heart disease and diabetes (hdd) prediction | |
US20220019892A1 (en) | Dialysis event prediction | |
Arefeen et al. | Glysim: Modeling and simulating glycemic response for behavioral lifestyle interventions | |
Fayaz et al. | Multi-Scale and Context Aware Optimized Glucose Prediction using Neural Networks |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
STPP | Information on status: patent application and granting procedure in general |
Free format text: DOCKETED NEW CASE - READY FOR EXAMINATION |
|
AS | Assignment |
Owner name: NEC CORPORATION, JAPAN Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNORS:NI, JINGCHAO;CHENG, WEI;CHEN, HAIFENG;AND OTHERS;REEL/FRAME:065133/0378 Effective date: 20220325 Owner name: NEC LABORATORIES AMERICA, INC., NEW JERSEY Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNORS:NI, JINGCHAO;CHENG, WEI;CHEN, HAIFENG;AND OTHERS;REEL/FRAME:065133/0378 Effective date: 20220325 |