CN111640481B - Parkinson's disease drug recommendation model based on multi-source data fusion - Google Patents

Parkinson's disease drug recommendation model based on multi-source data fusion Download PDF

Info

Publication number
CN111640481B
CN111640481B CN202010443048.5A CN202010443048A CN111640481B CN 111640481 B CN111640481 B CN 111640481B CN 202010443048 A CN202010443048 A CN 202010443048A CN 111640481 B CN111640481 B CN 111640481B
Authority
CN
China
Prior art keywords
symptoms
motor
drug
symptom
representation
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.)
Active
Application number
CN202010443048.5A
Other languages
Chinese (zh)
Other versions
CN111640481A (en
Inventor
史颖欢
陈震涛
高阳
张丽
潘杨
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Jiangsu Wanwei Aisi Network Intelligent Industry Innovation Center Co ltd
Nanjing University
Original Assignee
Jiangsu Wanwei Aisi Network Intelligent Industry Innovation Center Co ltd
Nanjing University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Jiangsu Wanwei Aisi Network Intelligent Industry Innovation Center Co ltd, Nanjing University filed Critical Jiangsu Wanwei Aisi Network Intelligent Industry Innovation Center Co ltd
Priority to CN202010443048.5A priority Critical patent/CN111640481B/en
Publication of CN111640481A publication Critical patent/CN111640481A/en
Application granted granted Critical
Publication of CN111640481B publication Critical patent/CN111640481B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

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
    • G16H20/10—ICT specially adapted for therapies or health-improving plans, e.g. for handling prescriptions, for steering therapy or for monitoring patient compliance relating to drugs or medications, e.g. for ensuring correct administration to patients
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61P—SPECIFIC THERAPEUTIC ACTIVITY OF CHEMICAL COMPOUNDS OR MEDICINAL PREPARATIONS
    • A61P25/00—Drugs for disorders of the nervous system
    • A61P25/14—Drugs for disorders of the nervous system for treating abnormal movements, e.g. chorea, dyskinesia
    • A61P25/16—Anti-Parkinson drugs
    • 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

Landscapes

  • Health & Medical Sciences (AREA)
  • Engineering & Computer Science (AREA)
  • Biomedical Technology (AREA)
  • Public Health (AREA)
  • General Health & Medical Sciences (AREA)
  • Medical Informatics (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Neurology (AREA)
  • Neurosurgery (AREA)
  • Epidemiology (AREA)
  • Medicinal Chemistry (AREA)
  • Primary Health Care (AREA)
  • Chemical & Material Sciences (AREA)
  • Pathology (AREA)
  • Databases & Information Systems (AREA)
  • Data Mining & Analysis (AREA)
  • Psychology (AREA)
  • Chemical Kinetics & Catalysis (AREA)
  • General Chemical & Material Sciences (AREA)
  • Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
  • Organic Chemistry (AREA)
  • Pharmacology & Pharmacy (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Animal Behavior & Ethology (AREA)
  • Veterinary Medicine (AREA)
  • Medical Treatment And Welfare Office Work (AREA)

Abstract

The invention provides a medicine recommendation model based on multi-source data fusion and particularly aims at parkinsonism, and belongs to the field of computer application. A Parkinson's disease drug recommendation model based on multi-source data fusion comprises the following steps: step (1) early data collection of motor and non-motor symptoms of PD patients and corresponding doctors prescribe clinical treatment, step (2) feature-based and similarity-based representations of the design of the observed motor and non-motor symptoms; step (3) learning a potential symptom space that relates the observed symptom to the prescribed medication; step (4) iteratively optimizing PALAS model parameters; step (5) saving model parameters; step (6) test results of the test set are evaluated. The invention is implemented by analyzing observed motor and non-motor symptoms of PD patients, and a computer automatically predicts drugs suitable for Parkinson patients, thereby providing reference for neuropathologists from a data-driven perspective.

Description

Parkinson's disease drug recommendation model based on multi-source data fusion
Technical Field
The invention relates to a parkinsonism drug recommendation model based on multi-source data fusion, and belongs to the field of computer application.
Background
Parkinson's Disease (PD) is a chronically developed neurological disorder with symptoms such as difficulty in activity and sleep, tremors, dizziness and syncope. It is now considered the second most common neurodegenerative disease, suffering from PD in nearly one million people in the united states. Although studies have shown that parkinson's disease may be related to genetic and environmental factors, the exact cause of parkinson's disease is still unknown.
In recent years, computer-aided methods for PD patients have become a widespread research hotspot for better guidance of interventional therapies. All studies have focused on computer-aided diagnosis for predicting whether a patient belongs to PD early. No study has been conducted regarding the recommendation of a computer PD drug prescription, i.e., selecting an appropriate daily therapeutic drug for a PD patient based on the observed symptoms.
Prescription prediction tasks are clinically significant and viable in practice. In traditional patient-based prescription methods, the expertise of a neuropathologist is required to prescribe a drug for a PD patient. Unfortunately, there is virtually no universal standard for optimal prescriptions. Different neuropathologists often prescribe different therapeutic drugs based on their own experience and judgment. Furthermore, if other factors are considered, such as price and implicit side effects of the drug, it becomes more complex and difficult to find (near) optimal prescriptions for each patient. Thus, drug prescriptions generated by combining data-driven machine learning techniques with patient-based prescriptions are very useful and can be used as a guideline or reference, especially when there are multiple perspectives between different neuropathologists.
In recent years, computer-aided PD diagnosis work is continually evolving. Fang et al propose a method for multi-modal consensus representation learning for image classification that proposes maximizing a difference metric to improve generalization ability. For the image annotation task, zou et al developed a method called MVML by fusing multiple single-label learners in an enhanced framework, where the base learner is an improved multi-modal based SVM. However, if the two methods are directly used for research, the predictive performance may be very limited because the inherent association of symptoms with drugs is not considered and the sparse nature of the drug label matrix is ignored.
Disclosure of Invention
The invention provides a drug recommendation model based on multi-source data fusion, which is specially aimed at parkinsonism, and aims at predicting a proper therapeutic drug according to observed motion and non-motion symptoms of parkinsonism patients.
A Parkinson's disease drug recommendation model based on multi-source data fusion comprises the following steps:
the method comprises the following steps of (1) collecting movement and non-movement symptoms of PD patients by early-stage data and prescribing medicines for clinical treatment by corresponding doctors;
step (2) feature-based and similarity-based representations of designs based on observed motor and non-motor symptoms;
step (3) learning a potential symptom space that relates the observed symptom to the prescribed medication;
step (4) iteratively optimizing PALAS model parameters;
step (5) saving model parameters;
in a further embodiment, the data collection of step (1) includes motor and non-motor symptoms of 136 PD patients between the ages of 48 and 92 years and the experienced neuropathologist provides a drug regimen appropriate for clinical treatment for each patient based on the symptoms.
In a further embodiment, the data representation in step (2) is a multimodal representation, a feature-based representation representing observed motor and non-motor symptoms is designed and a focused similarity-based representation is designed to reflect patient-aspect similarities using various kernel functions, and the feature-based and similarity-based representations are then combined into a multimodal representation of observed motor and non-motor symptoms.
In a further embodiment, the potential symptom space learning in step (3) for associating the observed symptom with the prescription drug is to describe the complex relationship between the observed symptom and the therapeutic drug, and it is more feasible and reasonable to use the inherent symptom-drug association by multi-label learning to jointly learn all the prediction tasks and introduce the potential symptom space, so that the overall prediction performance can be improved; by learning the potential symptom space, two transformations are actually learned together, (1) the observed symptom is transformed into a potential symptom transformation, and (2) the potential symptom is transformed into a therapeutic drug. The main considerations for learning these two transitions are as follows:
1) The observed conversion of symptoms to underlying symptoms; the potential symptom space is used to fully fuse information from different modalities and should well represent all modalities in a multi-modal representation, with each individual modality transformed, the learned potential symptom representation should have a relatively small F-norm distance;
2) Treating potential symptoms of drug conversion; the transformation matrix mapping the potential symptom representations to the drug labels should be sparse, as each prescription drug can only treat a limited number of symptoms.
Given a multi-modal representation of all samples (6 modalities)
Figure GDA0004219383040000021
Real drug tag matrix of n training samples +.>
Figure GDA0004219383040000022
Wherein->
Figure GDA0004219383040000023
c is the total number of drugs; the goal is to predict drug labels for m test samples; to keep the dimensions consistent +.>
Figure GDA0004219383040000024
As an additional drug class label matrix, where Y last m columns are set to 0 (using only the observation features of the training samples) since the test samples are not available during the training phase;
the PALAS model was developed to accomplish this predictive task, in which a potential symptom representation was learned
Figure GDA0004219383040000031
To correlate the observed symptoms with prescribed medications, where k represents the number of potential symptoms. The main objective function of PALAS is mathematically defined as follows:
Figure GDA0004219383040000032
wherein the method comprises the steps of
Figure GDA0004219383040000033
Is to observe a latent symptom transformation matrix to ensure that the latent symptom space P can represent the ith modality (X i )。/>
Figure GDA0004219383040000034
Is a transition matrix from symptoms to drugs that is used to relate different potential symptoms to different drugs. V is limited to sparsity because one drug can only alleviate very few symptoms, and from a clinical point of view, few patients take all possible drugs; />
Figure GDA0004219383040000035
Is a binary diagonal matrix that can be expressed as:
Figure GDA0004219383040000036
for representing training samples in a push-through setting, so that only training samples will participate in the prediction loss; thus, while the test sample itself does not aid in drug label prediction, combining the test sample can aid in learning P and U i Thus solving the problem of limited samples in the study; alpha and beta are regularization parameters for different terms in the equalization equation.
In a further embodiment, the PALAS model in step (4) employs an alternating optimization strategy; (1) fix P and V, update U i The method comprises the steps of carrying out a first treatment on the surface of the (2) Fixed U i And P, updating V; (3) fixed U i And V, updating P.
In a further embodiment, the given test sample z in step (6) may be defined by
Figure GDA0004219383040000037
Indicating, drug label->
Figure GDA0004219383040000038
The method can be obtained by the following steps:
Figure GDA0004219383040000039
the beneficial effects are that:
(1) The present invention proposes a new model of a latent symptom learning based prescription system (PrescriptionviALearninglAtentSymptom, PALAS) that predicts prescriptions using a multi-modal representation of data by integrating a feature-based representation and a similarity-based representation to design the multi-modal representation; the recommendation model may be implemented by analyzing observed motor and non-motor symptoms of PD patients (motor symptoms reflect a state of reduced motor skills, while non-motor symptoms describe a state of cognitive ability), and a computer automatically predicts drugs appropriate for Parkinson's Disease (PD) patients, thereby providing a reference to a neuropathologist from a data-driven perspective.
(2) By predicting the data collected in PD patients, the accuracy of the present invention reaches 87.5% over other competing methods. The invention has extremely high effectiveness and clinical potential.
Drawings
FIG. 1 is a diagram of a method construction of the present invention;
FIG. 2 is a diagram of a PALAS model framework in accordance with the present invention;
FIG. 3 is a graph of the results of the proposed method and the existing method on different evaluation indicators;
FIG. 4 is a graph comparing clinical accuracy of prescription predictions using different methods.
The specific embodiment is as follows:
the invention will be described in further detail with reference to the drawings and specific embodiments thereof, for the purpose of showing in detail the objects, features, and advantages of the present invention.
Prescription prediction tasks are clinically significant and viable in practice. In traditional patient-based prescription methods, the expertise of a neuropathologist is required to prescribe a drug for a PD patient. Unfortunately, there is virtually no universal standard for optimal prescriptions. Different neuropathologists often prescribe different therapeutic drugs based on their own experience and judgment. Furthermore, if other factors are considered, such as price and implicit side effects of the drug, it becomes more complex and difficult to find (near) optimal prescriptions for each patient. Thus, drug prescriptions generated by combining data-driven machine learning techniques with patient-based prescriptions are very useful and can be used as a guideline or reference, especially when there are multiple perspectives between different neuropathologists.
The core of the invention comprises the following parts:
(1) PD study data collection for training
(2) Multi-modal representation based on features and similarity
(3) PALAS model latent symptom space learning
(4) PALAS alternate optimization strategy
1. PD study data collection available for training:
exercise and non-exercise symptoms of 136 PD patients were recorded as training data. Motor symptoms record the daily movements of a subject, whereas non-motor symptoms describe the mental state of the subject, e.g., impression, mind, etc. All of these PD patients also improved symptoms after clinical treatment, as well as the drugs used in the continuous treatment. By taking patient-based prescriptions as real prescriptions, this information can be used to evaluate the performance of computer-aided prescriptions.
In this study, a total of 31 drugs including pramipexole dihydrochloride, spinoceram, alprazolam, carbidopa, vitamin E, donepezil, zopiclone, coenzyme Q10, ai Peirui pine hydrochloride, etc. were used, which have been shown to have an effective effect in improving PD in the previous neuroscience study. To better record this information, a binary matrix Y is used as a drug label matrix to represent the relationship between the patient and the drug. In particular, if and only if the ith patient takes the jth medication，Y ij =1, otherwise Y ij ＝0。
In addition, 55 motor symptoms and 143 non-motor symptoms were also observed and recorded in this study.
Motor symptoms. Motor symptoms often reflect whether a patient has problems with reduced motor skills in daily activities. Based on clinical observations, medical measurements and necessary quality assurance, neuropathologists rank symptoms according to a predefined digital scale.
Non-motor symptoms. Unlike motor symptoms, non-motor symptoms do not describe motor, coordination, physical tasks or activity abilities, but rather record information such as the cognitive abilities, mental states and physical conditions of the subject.
2. Multimodal representation based on features and similarity:
order the
Figure GDA0004219383040000051
And->
Figure GDA0004219383040000052
Feature-based representations of observed motor and non-motor symptoms, respectively, where n and m represent the number of exercises and the number of test samples, respectively (i.e., number of subjects, d 1 And d 2 Indicating the number of motor and non-motor symptoms, respectively. From the collected data, d 1 ＝55，d 2 ＝143.
In addition, several similarity-based representations have been designed to reflect patient-related similarities. In particular, various kernels are utilized, such as a linear kernel, a Gaussian kernel, and χ 2 The chi-square kernel calculates the similarity between any two feature vectors,
Figure GDA0004219383040000053
i.e. feature matrix X 1 And X 2 Is x= [ X ] 1 ,X 2 ]. The final similarity-based representation may be marked +.>
Figure GDA0004219383040000054
The feature-based and similarity-based representations are then combined into a multimodal representation of the observed motor and non-motor symptoms. In general, the multimodal representation of symptoms observed in PD patients is
Figure GDA0004219383040000055
Wherein d is i Is the dimension of the ith modality and s is the total number of different modalities. In this patent, s is set to 6 (i.e., 2 feature-based and 4 similarity-based representations).
Palas model symptom space learning:
the potential symptom space learning for associating the observed symptoms with the prescription drugs is to describe the complex relationship between the observed symptoms and the therapeutic drugs, and it is more feasible and reasonable to use the inherent symptom-drug association by learning all prediction tasks in combination through multi-label learning and introducing the potential symptom space, so that the overall prediction performance can be improved. Specifically, by learning the potential symptom space, two kinds of transformations are actually learned together, namely, (1) the observed symptom is transformed into a potential symptom transformation, and (2) the potential symptom is transformed into a therapeutic drug. The main considerations for learning these two transitions are as follows:
1) The observed conversion of symptoms to latent symptoms. The potential symptom space is used to fully fuse information from different modalities and should represent all modalities well in a multi-modal representation. In other words, as each individual modality is transformed, the learned underlying symptom representation should have a relatively small F-norm distance.
2) Treating potential symptoms of drug conversion. The transformation matrix mapping the potential symptom representations to the drug labels should be sparse, as each prescription drug can only treat a limited number of symptoms.
Given a multi-modal representation of all samples (6 modalities)
Figure GDA0004219383040000061
Real drug tag matrix of n training samples +.>
Figure GDA0004219383040000062
Wherein->
Figure GDA0004219383040000063
c is the total number of drugs. The goal is to predict drug labels for m test samples. To keep the dimensions consistent +.>
Figure GDA0004219383040000064
As an additional drug class label matrix, where the last m columns of Y are set to 0 (i.e., using only the observed features of the training sample) since the test sample is not available during the training phase.
The PALAS model was proposed to accomplish this predictive task in which a potential symptom representation was learned
Figure GDA0004219383040000065
To correlate the observed symptoms with prescribed medications, where k represents the number of potential symptoms. The main objective function of PALAS is mathematically defined as follows:
Figure GDA0004219383040000066
wherein the method comprises the steps of
Figure GDA0004219383040000067
Is to observe a latent symptom transformation matrix to ensure that the latent symptom space P can represent the ith modality (i.e., X i )。/>
Figure GDA0004219383040000068
Is a transition matrix from symptoms to drugs that is used to relate different potential symptoms to different drugs. V is limited to sparsity because one drug can only alleviate very few symptoms. Moreover, from a clinical point of view, few patients take all possible medications. />
Figure GDA0004219383040000069
Is a binary diagonal matrix that can be expressed as:
Figure GDA00042193830400000610
it is used to represent training samples in a straight-through setting, so that only training samples will participate in the prediction loss. Thus, while the test sample itself does not aid in drug label prediction, combining the test sample can aid in learning P and U i Thus, the problem of limited samples in the study is solved. Alpha and beta are regularization parameters for different terms in the equalization equation.
PALAS alternate optimization strategy
(1) Fix P and V, update U i
(2) Fixed U i And P, update V
(3) Fixed U i And V, update P
As shown in figure 1, the invention provides a medicine recommendation model based on multi-source data fusion, which is specially provided for parkinsonism. The model training stage comprises the following specific steps:
step (1): the early data collection PD patients' motor and non-motor symptoms and the corresponding doctors prescribe clinical treatment. Exercise and non-exercise symptoms of 136 PD patients were recorded as training data. Motor symptoms record the daily movements of a subject, whereas non-motor symptoms describe the mental state of the subject, e.g., impression, mind, etc. All of these PD patients also improved symptoms after clinical treatment, as well as the drugs used in the continuous treatment.
Step (2): as shown in fig. 2, the design based on observed motor and non-motor symptoms is based on a feature-based and similarity-based representation. Including feature-based representations of observed motor and non-motor symptoms and focused similarity-based representations designed using various kernel functions to reflect patient-aspect similarity, and then combined into a multi-modal representation of observed motor and non-motor symptoms.
Step (3): as shown in fig. 2, a potential symptom space is learned that relates the observed symptoms to the prescribed medication. Given a multi-modal representation of all samples (6 modalities)
Figure GDA0004219383040000071
Real drug tag matrix of n training samples +.>
Figure GDA0004219383040000072
Wherein->
Figure GDA0004219383040000073
c is the total number of drugs. The goal is to predict drug labels for m test samples. To keep the dimensions consistent +.>
Figure GDA0004219383040000074
As an additional drug class label matrix, where the last m columns of Y are set to 0 (i.e., using only the observed features of the training sample) since the test sample is not available during the training phase.
The PALAS model was proposed to accomplish this predictive task in which a potential symptom representation was learned
Figure GDA0004219383040000075
To correlate the observed symptoms with prescribed medications, where k represents the number of potential symptoms. The main objective function of PALAS is mathematically defined as follows:
Figure GDA0004219383040000076
wherein the method comprises the steps of
Figure GDA0004219383040000077
Is to observe a latent symptom transformation matrix to ensure that the latent symptom space P can represent the ith modality (i.e., X i )。/>
Figure GDA0004219383040000078
Is a transition matrix from symptoms to drugs that is used to relate different potential symptoms to different drugs. V is limited to sparsity because one drug can only alleviate very few symptoms. Moreover, from a clinical point of view, few patients take all possible medications. />
Figure GDA0004219383040000079
Is a binary diagonal matrix that can be expressed as:
Figure GDA00042193830400000710
it is used to represent training samples in a straight-through setting, so that only training samples will participate in the prediction loss. Thus, while the test sample itself does not aid in drug label prediction, combining the test sample can aid in learning P and U i Thus, the problem of limited samples in the study is solved. Alpha and beta are regularization parameters for different terms in the equalization equation.
Step (4): iterative optimization of PALAS model parameters.
Step (5): and saving the model parameters.
Step (6): test set test results were tested and evaluated. A given test sample z may be defined by
Figure GDA0004219383040000081
Indicating, drug label->
Figure GDA0004219383040000082
The method can be obtained by the following steps:
Figure GDA0004219383040000083
the evaluation method proceeds from two aspects, respectively. One is the evaluation index hamming loss, one-error, coverage and ordering error used so far, as shown in fig. 3, is intended to evaluate multi-label predictions. These indices are used to provide a reference at the data driven point of view. On the other hand, the prayer reports the evaluation results from a clinical point of view, as shown in fig. 4, and in particular, three experienced neuropathologists scrutinized the prediction results. That is, a neuropathologist is required to determine from a clinical perspective whether a predicted drug can improve the observed symptoms of a patient. More than two-thirds of the neuropathologists agree that the prescription prediction is correct, otherwise it is considered incorrect. This assessment enables the calculation of overall clinical accuracy. For predictions for 136 patients, PALAS reached 87.5% accuracy (119 correct and 17 incorrect), exceeding all of the comparative methods SMBR, BR, CLR, ML-kNN, rank-SVM, MLVML-MM and MVML).

Claims (4)

1. A Parkinson's disease drug recommendation model based on multi-source data fusion comprises the following steps:
the method comprises the following steps of (1) collecting movement and non-movement symptoms of PD patients by early-stage data and prescribing medicines for clinical treatment by corresponding doctors;
step (2) designing feature-based and similarity-based representations based on the observed motor and non-motor symptoms, respectively: feature-based representation of motor and non-motor symptoms and use of linear kernels, gaussian kernels and χ 2 The chi-square kernel calculates a similarity-based representation of the similarity between any two feature vectors;
step (3) learning a potential symptom space which relates the observed symptom to the prescription drug, wherein the potential symptom space is learned by optimizing a designed objective function;
step (4) iteratively optimizing PALAS model parameters;
step (5) saving model parameters;
step (6) testing results of the testing set and evaluating;
the data representation in step (2) is a multimodal representation, a feature-based representation representing observed motor and non-motor symptoms is designed and a focused similarity-based representation is designed to reflect patient-aspect similarity using various kernel functions, and the feature-based and similarity-based representations are then combined into a multimodal representation of observed motor and non-motor symptoms;
the learning in the step (3) is to use the inherent symptom and medicine association by learning all prediction tasks in a multi-label learning combination way and introducing the potential symptom space; by learning the potential symptom space, two transformations are learned together, (1) the observed symptom is transformed into a potential symptom transformation, (2) the potential symptom is transformed into a therapeutic drug;
multi-modal representation of given all samples
Figure FDA0004219383030000011
Wherein d is i Is the dimension of the ith modality and the real drug label matrix of n training samples +.>
Figure FDA0004219383030000012
Wherein->
Figure FDA0004219383030000013
c is the total number of drugs; the goal is to predict drug labels for m test samples; to keep the dimensions consistent, introduce +.>
Figure FDA0004219383030000014
As an additional drug class mark matrix, and setting the last m columns of the class mark matrix Y to 0;
potential symptom representation learned in the PALAS model
Figure FDA0004219383030000015
Wherein k represents the number of potential symptoms; the objective function of PALAS is mathematically defined as follows:
Figure FDA0004219383030000016
wherein the method comprises the steps of
Figure FDA0004219383030000017
Is the observation of a latent symptom transformation matrix, the latent symptom space P can represent the ith modality (X i )，
Figure FDA0004219383030000018
Is a transition matrix from symptoms to drugs, V is limited to sparsity; />
Figure FDA0004219383030000019
Is a binary diagonal matrix expressed as:
Figure FDA0004219383030000021
alpha and beta are regularization parameters for different terms in the equalization equation.
2. The multi-source data fusion-based parkinsonism drug recommendation model according to claim 1, wherein: during the data collection of step (1), including motor and non-motor symptoms in 136 PD patients between the ages of 48 and 92 and experienced neuropathologists, provide a drug regimen appropriate for clinical treatment for each patient based on the symptoms.
3. The multi-source data fusion-based parkinsonism drug recommendation model according to claim 1, wherein: the PALAS model in the step (4) adopts an alternate optimization strategy; (1) fix P and V, update U i The method comprises the steps of carrying out a first treatment on the surface of the (2) Fixed U i And P, updating V; (3) fixed U i And V, updating P.
4. The multi-source data fusion-based parkinsonism drug recommendation model according to claim 1, wherein: the given test sample z in the step (6) is defined by
Figure FDA0004219383030000022
Representing, where s is the total number of different modalities, drug label +.>
Figure FDA0004219383030000023
Obtained by:
Figure FDA0004219383030000024
CN202010443048.5A 2020-05-22 2020-05-22 Parkinson's disease drug recommendation model based on multi-source data fusion Active CN111640481B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010443048.5A CN111640481B (en) 2020-05-22 2020-05-22 Parkinson's disease drug recommendation model based on multi-source data fusion

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010443048.5A CN111640481B (en) 2020-05-22 2020-05-22 Parkinson's disease drug recommendation model based on multi-source data fusion

Publications (2)

Publication Number Publication Date
CN111640481A CN111640481A (en) 2020-09-08
CN111640481B true CN111640481B (en) 2023-06-20

Family

ID=72333264

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010443048.5A Active CN111640481B (en) 2020-05-22 2020-05-22 Parkinson's disease drug recommendation model based on multi-source data fusion

Country Status (1)

Country Link
CN (1) CN111640481B (en)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117214398B (en) * 2023-09-04 2024-05-14 江苏省连云港环境监测中心 Deep underground water body pollutant detection method and system

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108717862A (en) * 2018-04-10 2018-10-30 四川骏逸富顿科技有限公司 A kind of careful square evolution model of the intelligence based on machine learning
CN109243567A (en) * 2018-08-14 2019-01-18 山东科技大学 A kind of drug recommended method excavated based on prescription data
CN110570903A (en) * 2019-08-12 2019-12-13 中山大学孙逸仙纪念医院 Medicine for improving activity of mesocerebral substantia nigra dopamine neurons and preventing and treating Parkinson's disease

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108717862A (en) * 2018-04-10 2018-10-30 四川骏逸富顿科技有限公司 A kind of careful square evolution model of the intelligence based on machine learning
CN109243567A (en) * 2018-08-14 2019-01-18 山东科技大学 A kind of drug recommended method excavated based on prescription data
CN110570903A (en) * 2019-08-12 2019-12-13 中山大学孙逸仙纪念医院 Medicine for improving activity of mesocerebral substantia nigra dopamine neurons and preventing and treating Parkinson's disease

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
基于医疗过程挖掘与患者体征的药物推荐方法;李鹏飞等;《计算机集成制造系统》;20200319;全文 *

Also Published As

Publication number Publication date
CN111640481A (en) 2020-09-08

Similar Documents

Publication Publication Date Title
Egger et al. Medical deep learning—A systematic meta-review
Loh et al. Application of deep learning models for automated identification of Parkinson’s disease: A review (2011–2021)
Zhao et al. Diagnosis of autism spectrum disorders using multi-level high-order functional networks derived from resting-state functional mri
Brand et al. Joint multi-modal longitudinal regression and classification for alzheimer’s disease prediction
Lima et al. A comprehensive survey on the detection, classification, and challenges of neurological disorders
Robinson et al. Defining phenotypes from clinical data to drive genomic research
Aghili et al. Predictive modeling of longitudinal data for alzheimer’s disease diagnosis using rnns
Chowriappa et al. Introduction to machine learning in healthcare informatics
ElNakieb et al. The role of diffusion tensor MR imaging (DTI) of the brain in diagnosing autism spectrum disorder: promising results
Odusami et al. Explainable deep-learning-based diagnosis of Alzheimer’s disease using multimodal input fusion of PET and MRI Images
Wan et al. An artificial intelligent risk classification method of high myopia based on fundus images
Zhou et al. A survey of deep learning for alzheimer’s disease
Zhu et al. Statistical learning methods for neuroimaging data analysis with applications
CN111640481B (en) Parkinson's disease drug recommendation model based on multi-source data fusion
Keles et al. The past, current, and future of neonatal intensive care units with artificial intelligence: a systematic review
Tassew et al. A Comprehensive Review of the Application of Machine Learning in Medicine and Health Care
Tanveer et al. Ensemble deep learning for Alzheimer’s disease characterization and estimation
Olaniyan et al. Prevention and diagnosis of neurodegenerative diseases using machine learning models
Zhou et al. Attentive deep canonical correlation analysis for diagnosing alzheimer’s disease using multimodal imaging genetics
Duan et al. Deep learning based multimodal biomedical data fusion: An overview and comparative review
Bhandarkar et al. Deep learning based computer aided diagnosis of Alzheimer’s disease: a snapshot of last 5 years, gaps, and future directions
Zhang et al. Value of rehabilitation training for children with cerebral palsy diagnosed and analyzed by computed tomography imaging information features under deep learning
Zhang et al. Landscape of big medical data: a pragmatic survey on prioritized tasks
Wassan et al. Role of deep learning in predicting aging-related diseases: a scoping review
Turner et al. The classification of movement in infants for the autonomous monitoring of neurological development

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant