CN116434908A - Method and device for quantitatively and hierarchically evaluating bradykinesia based on time convolution network - Google Patents
Method and device for quantitatively and hierarchically evaluating bradykinesia based on time convolution network Download PDFInfo
- Publication number
- CN116434908A CN116434908A CN202310701314.3A CN202310701314A CN116434908A CN 116434908 A CN116434908 A CN 116434908A CN 202310701314 A CN202310701314 A CN 202310701314A CN 116434908 A CN116434908 A CN 116434908A
- Authority
- CN
- China
- Prior art keywords
- bradykinesia
- kinematic data
- target object
- data
- layer
- 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
- 206010006100 Bradykinesia Diseases 0.000 title claims abstract description 135
- 208000006083 Hypokinesia Diseases 0.000 title claims abstract description 135
- 238000000034 method Methods 0.000 title claims abstract description 103
- 238000011158 quantitative evaluation Methods 0.000 claims abstract description 29
- 238000012549 training Methods 0.000 claims abstract description 13
- 208000037170 Delayed Emergence from Anesthesia Diseases 0.000 claims description 45
- 230000008569 process Effects 0.000 claims description 37
- 238000010606 normalization Methods 0.000 claims description 36
- 238000012545 processing Methods 0.000 claims description 34
- 238000001914 filtration Methods 0.000 claims description 24
- 238000007781 pre-processing Methods 0.000 claims description 13
- 230000006870 function Effects 0.000 claims description 12
- 230000033001 locomotion Effects 0.000 claims description 12
- 230000001364 causal effect Effects 0.000 claims description 6
- 238000004590 computer program Methods 0.000 claims description 5
- 238000011156 evaluation Methods 0.000 abstract description 18
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 25
- 201000010099 disease Diseases 0.000 description 21
- 238000013139 quantization Methods 0.000 description 19
- 230000003925 brain function Effects 0.000 description 12
- 238000010586 diagram Methods 0.000 description 12
- 210000001364 upper extremity Anatomy 0.000 description 12
- 210000004556 brain Anatomy 0.000 description 11
- 238000013210 evaluation model Methods 0.000 description 10
- 230000001133 acceleration Effects 0.000 description 8
- 210000004247 hand Anatomy 0.000 description 8
- 230000009471 action Effects 0.000 description 5
- 238000004891 communication Methods 0.000 description 5
- 230000000694 effects Effects 0.000 description 5
- 208000035475 disorder Diseases 0.000 description 4
- 208000012661 Dyskinesia Diseases 0.000 description 3
- 238000013461 design Methods 0.000 description 3
- 238000003745 diagnosis Methods 0.000 description 3
- 210000003414 extremity Anatomy 0.000 description 3
- 238000007476 Maximum Likelihood Methods 0.000 description 2
- 230000004913 activation Effects 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 238000005315 distribution function Methods 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 210000002569 neuron Anatomy 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 230000001020 rhythmical effect Effects 0.000 description 2
- 210000000323 shoulder joint Anatomy 0.000 description 2
- WHXSMMKQMYFTQS-UHFFFAOYSA-N Lithium Chemical compound [Li] WHXSMMKQMYFTQS-UHFFFAOYSA-N 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000013480 data collection Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 238000004880 explosion Methods 0.000 description 1
- 239000012634 fragment Substances 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 229910052744 lithium Inorganic materials 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 230000002787 reinforcement Effects 0.000 description 1
- 238000004171 remote diagnosis Methods 0.000 description 1
- 230000000284 resting effect Effects 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 208000024891 symptom Diseases 0.000 description 1
- 210000000707 wrist Anatomy 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
- G16H20/30—ICT specially adapted for therapies or health-improving plans, e.g. for handling prescriptions, for steering therapy or for monitoring patient compliance relating to physical therapies or activities, e.g. physiotherapy, acupressure or exercising
-
- 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
- G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
-
- 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
- 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
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/30—Computing systems specially adapted for manufacturing
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biophysics (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Public Health (AREA)
- General Physics & Mathematics (AREA)
- Medical Informatics (AREA)
- Artificial Intelligence (AREA)
- Computational Linguistics (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Epidemiology (AREA)
- Primary Health Care (AREA)
- Pathology (AREA)
- Databases & Information Systems (AREA)
- Physical Education & Sports Medicine (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention provides a method and a device for quantitatively and hierarchically evaluating bradykinesia based on a time convolution network, which are applied to the field of medical care informatics, wherein the method comprises the following steps: acquiring kinematic data of a target object; obtaining a bradykinesia grade of the target object based on the kinematic data and the trained bradykinesia quantitative evaluation model; the bradykinesia quantitative evaluation model is obtained by training a time convolution network according to a sample kinematics data set with a bradykinesia grade label. Therefore, quantitative evaluation is objectively carried out on the bradykinesia degree of the target object, and the accuracy of the evaluation result is improved.
Description
Technical Field
The invention relates to the technical field of medical care informatics, in particular to a method and a device for quantitatively and hierarchically evaluating bradykinesia based on a time convolution network.
Background
In the prior art, the degree of bradykinesia is usually evaluated subjectively by an experienced clinician according to the behavior of the object to be evaluated, which is disadvantageous in that the quality of the evaluation result has high degree of dependence on the experience level of the clinician, and the evaluation result cannot objectively and quantitatively describe the change of the degree of dyskinesia of the object to be evaluated.
Disclosure of Invention
The invention provides a quantitative grading evaluation method and device for bradykinesia based on a time convolution network, which are used for solving the defects that the quality of an evaluation result in the prior art has higher degree of dependence on the experience level of a clinician and the evaluation result cannot objectively and quantitatively describe the change of the dyskinesia degree of a patient, realizing the quantitative evaluation of the bradykinesia degree of a target object and improving the accuracy of the evaluation result.
The invention provides a method for quantitatively and hierarchically evaluating bradykinesia based on a time convolution network, which comprises the following steps:
acquiring kinematic data of a target object;
obtaining a bradykinesia grade of the target object based on the kinematic data and the trained bradykinesia quantitative evaluation model;
the bradykinesia quantitative evaluation model is obtained by training a time convolution network according to a sample kinematics data set with a bradykinesia grade label.
According to the bradykinesia quantization grading evaluation method based on the time convolution network, the bradykinesia quantization evaluation model comprises an input layer, a residual error network, a full connection layer and an output layer, wherein the input layer, the residual error network, the full connection layer and the output layer are sequentially connected;
the residual network comprises a plurality of residual blocks which are connected in sequence, and any two adjacent residual blocks are connected through a 1X 1 convolution layer.
According to the motion retardation quantization hierarchical assessment method based on the time convolution network, each residual block comprises two convolution units, and the structure of each convolution unit is sequentially connected with an expansion causal convolution layer, a weight normalization layer, a linear rectification function ReLU layer and a Dropout layer.
According to the method for quantitatively and hierarchically evaluating the bradykinesia based on the time convolution network, in the residual error network, the expansion factor of the residual error block connected with the input layer is 1, and the expansion factor of any residual error block is 2 times of the expansion factor of the previous residual error block.
According to the method for quantitatively and hierarchically evaluating the bradykinesia based on the time convolution network, which is provided by the invention, the bradykinesia grade of the target object is obtained based on the kinematic data and the trained quantitative evaluation model of the bradykinesia, and the method comprises the following steps:
preprocessing the kinematic data, inputting the preprocessed kinematic data into a trained quantitative evaluation model of the bradykinesia to obtain the bradykinesia grade of the target object;
the preprocessing comprises filtering processing and/or normalization processing;
wherein the filtering process includes:
obtaining frequency domain data corresponding to the time domain data by fast Fourier transform, and performing inverse fast Fourier transform after setting the amplitude corresponding to zero frequency and the amplitude corresponding to frequency higher than a preset frequency threshold in the frequency domain data;
the normalization is performed according to the following formula:
in the method, in the process of the invention,
kinematic data before normalization processing; />
Is the kinematic data after normalization treatment.
According to the method for quantitatively and hierarchically evaluating the bradykinesia based on the time convolution network, the sample kinematic data set is obtained according to the following mode:
acquiring a sample original kinematic data set;
performing one or more of filtering processing, data enhancement processing and normalization processing on the original kinematic data set of the sample to obtain the kinematic data set of the sample;
wherein the data enhancement processing includes:
expanding the sample original kinematic data set using a sliding window method; and/or the number of the groups of groups,
gaussian white noise is added to the sample raw kinematic dataset.
According to the method for quantitatively and hierarchically evaluating the bradykinesia based on the time convolution network, which is provided by the invention, the kinematic data of the target object comprises one or more of the following:
kinematic data when the target object performs a rest task;
kinematic data of the target object when performing the stretching task;
kinematic data when the target object performs the rotation task.
The invention also provides a device for quantitatively and hierarchically evaluating the bradykinesia based on the time convolution network, which comprises the following components:
the acquisition data module is used for acquiring the kinematic data of the target object;
the output grade module is used for obtaining the bradykinesia grade of the target object based on the kinematic data and the trained bradykinesia quantitative evaluation model;
the bradykinesia quantitative evaluation model is obtained by training a time convolution network according to a sample kinematics data set with a bradykinesia grade label.
The invention also provides an electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing a method for quantitative hierarchical assessment of bradykinesia based on a time convolution network as described in any one of the above.
The present invention also provides a non-transitory computer readable storage medium having stored thereon a computer program which, when executed by a processor, implements a method of quantitative hierarchical assessment of bradykinesia based on a time convolution network as described in any one of the above.
According to the method and the device for quantitatively and hierarchically evaluating the bradykinesia based on the time convolution network, the bradykinesia quantitative evaluation model is obtained through training the time convolution network according to the sample kinematic data set with the bradykinesia grade label, the kinematic data of the target object is obtained, and the bradykinesia grade of the target object is obtained based on the kinematic data and the trained bradykinesia quantitative evaluation model, so that quantitative evaluation is objectively carried out on the bradykinesia degree of the target object, and the accuracy of an evaluation result is improved.
Drawings
In order to more clearly illustrate the invention or the technical solutions of the prior art, the following description will briefly explain the drawings used in the embodiments or the description of the prior art, and it is obvious that the drawings in the following description are some embodiments of the invention, and other drawings can be obtained according to the drawings without inventive effort for a person skilled in the art.
FIG. 1 is a flow chart of a method for quantitatively and hierarchically assessing bradykinesia based on a time convolution network provided by the present invention;
FIG. 2 is a schematic diagram of a model for quantitative assessment of bradykinesia according to the present invention;
fig. 3 is a schematic structural diagram of a residual block according to the present invention;
fig. 4 is a schematic diagram of a portable lightweight evaluation system based on a wearable device and a cloud platform provided by the invention;
FIG. 5 is a schematic diagram of a specific structure of an evaluation model according to the present invention;
FIG. 6 is a schematic diagram of a device for quantitatively and hierarchically assessing bradykinesia based on a time convolution network according to the present invention;
fig. 7 is a schematic diagram of an entity structure of an electronic device according to the present invention.
Detailed Description
For the purpose of making the objects, technical solutions and advantages of the present invention more apparent, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings, and it is apparent that the described embodiments are some embodiments of the present invention, not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
Fig. 1 is a flow chart of a method for quantitatively and hierarchically evaluating bradykinesia based on a time convolution network according to the present invention, as shown in fig. 1, the method comprises the following steps:
And step 101, obtaining the bradykinesia grade of the target object based on the kinematic data and the trained bradykinesia quantitative evaluation model.
Wherein the bradykinesia quantitative assessment model is trained from a time convolution network from a sample kinematic data set with bradykinesia class labels.
Specifically, the execution subject of the method provided by the present invention may be a processing device that may receive input from other devices and has a certain computing capability, and the method provided by the present invention will be described below by taking a computer device as an example.
The target object is an object which needs to be subjected to the quantitative evaluation of the bradykinesia, and the computer equipment can acquire the kinematic data of the target object firstly.
In some embodiments, the target object may be worn with a wearable device deployed with an inertial sensor, and the wearable device deployed with the inertial sensor collects kinematic data of the target object and sends the kinematic data to the computer device. For example, the target object may be worn with a bracelet equipped with inertial sensors that may collect various kinematic data of the target object, such as triaxial acceleration, triaxial angular velocity, pitch and yaw angles of hand movement in real time, and the like.
The computer device may train the time convolution network to obtain a bradykinesia quantization assessment model based on the sample kinematic data with the bradykinesia class label, and then obtain the bradykinesia class of the target object based on the acquired kinematic data of the target object and the trained bradykinesia quantization assessment model.
Optionally, the wearable device may preprocess the collected original kinematic data, and then send the preprocessed kinematic data to the computer device, where the computer device inputs the preprocessed kinematic data into the trained quantization and evaluation model for bradykinesia to determine the bradykinesia level of the target object.
Optionally, deriving a bradykinesia level of the target object based on the kinematic data and the trained bradykinesia quantitative assessment model comprises:
preprocessing the kinematic data, and inputting the preprocessed kinematic data into a trained quantitative evaluation model of the bradykinesia to obtain the bradykinesia grade of a target object;
the preprocessing comprises filtering processing and/or normalization processing;
wherein the filtering process includes:
obtaining frequency domain data corresponding to the time domain data by fast Fourier transform of the time domain data of the kinematic data, and carrying out fast Fourier inverse transform after zeroing amplitude corresponding to zero frequency and amplitude corresponding to frequency higher than a preset frequency threshold in the frequency domain data;
the normalization process is performed according to the following formula:
in the method, in the process of the invention,
kinematic data before normalization processing; />
Is the kinematic data after normalization treatment.
Specifically, in the case where the acquired kinematic data is original kinematic data acquired by the wearable device, the computer device may perform preprocessing on the acquired kinematic data and then input a trained quantization evaluation model of bradykinesia, thereby obtaining a bradykinesia level of the target object.
The preprocessing of the kinematic data may be a filtering process of the kinematic data, or a normalizing process of the kinematic data, or both a filtering process and a normalizing process of the kinematic data.
The filtering process may be mapping the kinematic data to a corresponding frequency domain using a fast fourier transform method, and then zeroing an amplitude corresponding to a zero frequency and an amplitude corresponding to a frequency higher than a preset frequency threshold in the frequency domain. On the basis, the processed frequency domain data is inverted to time domain data through a fast Fourier transform function, so that the filtering effect is achieved.
For example, for a length N, a finite time series of dimension 2 is
The fast fourier transform formula of the sequence can be expressed as:
its inverse fast fourier transform can be expressed as:
in the method, in the process of the invention,
is a twiddle factor in the fast fourier transform.
In the process of data acquisition, the acceleration value at the initial moment can be influenced by the slight difference of the wearing positions of different target object wearing devices. The filtering process may eliminate the effect of this on the quality of the kinematic data acquired by the target object and on the performance of the model.
The normalization process may be performed according to the following formula:
in the method, in the process of the invention,
kinematic data before normalization processing; />
Is the kinematic data after normalization treatment. The training bradykinesia quantitative evaluation model is input after the normalization processing is carried out on the kinematic data, so that the recognition performance of the training bradykinesia quantitative evaluation model can be improved.
Alternatively, the sample kinematic data set may be obtained according to the following manner:
acquiring a sample original kinematic data set;
performing one or more of filtering processing, data enhancement processing and normalization processing on the original kinematic data set of the sample to obtain the kinematic data set of the sample;
wherein, the data enhancement processing includes:
expanding the original kinematic data set of the sample by using a sliding window method; and/or the number of the groups of groups,
gaussian white noise is added to the original kinematic data set of the sample.
Specifically, a sample original kinematic data set may be acquired first, and the sample original kinematic data set may also be obtained by acquiring original kinematic data of a plurality of objects through a wearable device.
One or more of the filtering, data enhancement and normalization processes may then be performed on the sample raw kinematic data set to obtain a sample kinematic data set for training the bradykinesia quantitative assessment model.
The filtering and normalization processes are similar to the preprocessing process for the kinematic data, and will not be described in detail here. The data enhancement processing of the original kinematic data set of the sample can be that the original kinematic data set of the sample is expanded by using a sliding window method; alternatively, gaussian white noise (Gaussian white noise) may be added to the sample raw kinematic data set; alternatively, both the expansion of the sample raw kinematic data set and the addition of gaussian white noise to the sample raw kinematic data set may be performed using a sliding window approach.
The time sequence fragments are intercepted by a sliding window method, so that the acquired original kinematic data of the sample can be remodeled into samples with specified lengths, and the model identification precision is improved. Preferably, the time window may be set to 5 seconds.
The gaussian white noise may be generated using a pseudo-random number generator, with the amount of noise added being a configurable super-parameter. Preferably, gaussian white noise randomly generated five times higher than the original kinematic data of the sample can be applied to the enhancement data as follows:
in the middle of
Representing kinematic data after adding white gaussian noise,/->
Representing kinematic data before adding white gaussian noise,/->
Representing a normal distribution function with a mean value of 0 and a standard deviation of 0.1.
In order to reduce the dimension of the input data and obtain a more accurate result, it is preferable that the triaxial acceleration is synthesized into uniaxial acceleration data, and the uniaxial acceleration data and the y-axis angular velocity which has the strongest correlation with the motion process are selected as the kinematic data of the input bradykinesia quantization assessment model.
According to the time convolution network-based bradykinesia quantization and grading evaluation method, the bradykinesia quantization and evaluation model is obtained through training the time convolution network according to the sample kinematics data set with the bradykinesia grade label, the kinematics data of the target object is obtained, and the bradykinesia grade of the target object is obtained based on the kinematics data and the trained bradykinesia quantization and evaluation model, so that quantitative evaluation is objectively carried out on the bradykinesia degree of the target object, and the accuracy of an evaluation result is improved.
Optionally, the bradykinesia quantization assessment model comprises an input layer, a residual error network, a full connection layer and an output layer, wherein the input layer, the residual error network, the full connection layer and the output layer are sequentially connected;
the residual network comprises a plurality of residual blocks which are connected in sequence, and any two adjacent residual blocks are connected through a convolution layer of 1 multiplied by 1.
Specifically, fig. 2 is a schematic structural diagram of a bradykinesia quantization assessment model provided by the present invention, and as shown in fig. 2, the bradykinesia quantization assessment model may include an input layer, a residual network, a full connection layer, and an output layer, where the residual network includes a plurality of residual blocks connected in sequence.
In using the bradykinesia quantization assessment model, the input data passes through the input layer, the plurality of residual blocks, the full connection layer, and the output layer in order.
The convolution layer of 1 multiplied by 1 can be applied between any two adjacent residual blocks, so that the data output length and the input length of the residual blocks are kept consistent, the information of the elapsed time is kept, the convergence process of the model can be accelerated, and the accuracy of model evaluation can be improved.
After the data reaches the residual network through the input layer, the residual network formed by a plurality of residual blocks can carry out convolution processing and transmission on the input data for many times, then the data is flattened into one-dimensional signals by the full-connection layer to reach the output layer, the output layer can adopt a Softmax nonlinear function as a classifier, the one-dimensional signals obtained through the full-connection layer can obtain probability distribution of various categories after the one-dimensional signals pass through the Softmax function, and then the category with the largest probability can be used as a prediction result according to the maximum likelihood estimation principle:
in the method, in the process of the invention,
probability of being a predicted category, +.>
Is the model's score for this category, N represents the number of categories.
In some embodiments, the number of neurons in the output layer is 4, so that four levels of bradykinesia can be obtained, normal, mild, moderate, and severe, respectively. The degree of bradykinesia of a target subject can be more precisely divided than in the current solutions, which mostly only classify bradykinesia symptoms.
Optionally, each residual block includes two convolution units, and each convolution unit has a structure that an expansion causal convolution layer, a weight normalization layer, a linear rectification function (Rectified Linear Unit, reLU) layer, and a Dropout layer are sequentially connected.
Specifically, fig. 3 is a schematic structural diagram of a residual block provided by the present invention, and as shown in fig. 3, each residual block includes two convolution units, and each convolution unit has a structure that an expansion causal convolution layer, a weight normalization layer, a ReLU layer, and a Dropout layer are sequentially connected.
More features in the input data are learned through an expansion causal convolution layer, so that the recognition accuracy of the model is improved.
Optionally, in the residual network, the expansion factor of the residual block connected with the input layer is 1, and the expansion factor of any residual block is 2 times of the expansion factor of the previous residual block.
For example, preferably, the model for quantized evaluation of bradykinesia provided by the present invention may comprise 6 residual blocks, and as the model deepens, the expansion factors of the first to sixth residual blocks take values of 1,2,4,8,16,32 in order.
Normalizing the network weight by a weight normalization layer, accelerating the training process and improving the generalization capability of the model; the nonlinear activation operation is carried out by the ReLU layer, so that the nonlinear fitting capacity of the neural network is improved, and the expression capacity of the model is enhanced; overfitting of the model network is prevented by the Dropout layer randomly discarding nodes.
Preferably, the Dropout factor of each Dropout layer takes a value of 0.3.
Optionally, the kinematic data of the target object includes one or more of:
(1) Kinematic data when the target object performs a resting task.
Wherein, the rest task refers to a state that the target object keeps the static posture basically unchanged.
For example, the target object can be made to wear a wearable device with an inertial sensor disposed thereon to maintain a sitting posture and put both arms on both side handrails of a chair, and kinematic data of the target object in this state can be acquired.
(2) Kinematic data of the target object when performing the stretching task.
Wherein, the stretching task refers to a state that the target object keeps the stretching posture basically unchanged when stretching the limb.
For example, a target subject may be wearing a wearable device with an inertial sensor deployed to hold his hands level up to the front of the body and level with the shoulder joints while facing his palms towards the ground, while maintaining a sitting posture.
(3) Kinematic data when the target object performs the rotation task.
The rotation task refers to a state that a target object keeps rhythmically changing the active posture of the limb when stretching the limb.
For example, the target object can wear the wearable device provided with the inertial sensor, and on the basis of keeping the sitting posture and lifting the hands horizontally at the front side of the body and the height of the shoulder joints, the hands do rapid rotation motion, namely the upper limbs do the maximum amplitude of supination and supination motions, and meanwhile, the palms of the hands face and deviate from the ground in a rhythmic manner.
The method for quantitatively and hierarchically evaluating the bradykinesia based on the time convolution network is further described by the embodiment of a specific application scene.
It should be noted that, the present invention is not a disease diagnosis method for obtaining a diagnosis result, and the bradykinesia grade obtained by the quantitative and hierarchical assessment method for bradykinesia based on a time convolution network according to the present invention cannot be used as a diagnosis result of a patient with a disease. Although the following embodiments of the specific application scenario use a brain functional disease patient as a target object, the method for quantitatively and hierarchically evaluating the bradykinesia of the upper limb of the brain functional disease patient provided by the present invention is not meant to be a method for quantitatively and hierarchically evaluating the bradykinesia of the upper limb of the brain functional disease patient provided by the present invention, which can be used for diagnosing the brain functional disease patient.
The embodiment provides a quantitative and hierarchical evaluation system for the upper limb bradykinesia of a brain functional disease patient based on a time convolution network, which comprises the following equipment and main methods:
1. design of wearable sensor
In order to be able to acquire the upper limb movement signals of patients with brain function disorders, a bracelet equipped with an inertial sensor is used. The bracelet is internally provided with an inertial sensor module, a low-power-consumption Bluetooth module, an STM32 microprocessor, a lithium battery and other key devices. The bracelet can collect 8 kinematic data of a patient, including triaxial acceleration, triaxial angular speed and pitch angle and yaw angle data of hand real-time motion. Fig. 4 is a schematic diagram of the portable lightweight evaluation system based on the wearable device and the cloud platform, and as shown in fig. 4, the device can transmit kinematic data of a patient to a mobile phone end through bluetooth, and the mobile phone end can communicate with a remote computer and a cloud terminal to realize a remote diagnosis function. In the process of collecting upper limb kinematic data of a patient with brain function diseases, the bracelet is worn on the wrist of the patient.
2. Design of data collection paradigm
Aiming at the grading of the upper limb bradykinesia of a patient with brain function diseases, the embodiment provides a set of data acquisition paradigm, and the specific task flow is as follows:
(1) And executing a rest task. The brain function disorder patient wears the wearable device to keep sitting and puts the arms on the two-sided armrest of the chair, and the action is kept for 15 seconds.
(2) Performing the stretching task. The brain function disease patient keeps sitting posture, lifts the hands horizontally on the front side of the body and at the same time makes the palms of the hands face the ground, and the action is kept for 15 seconds.
(3) And executing the rotation task. On the basis of the extension task, the brain functional disease patient does the rapid rotation action of the hands for 15 seconds, namely the maximum amplitude of supination and supination movement of the upper limbs, and simultaneously, the palms of the hands face and deviate from the ground in a rhythmic manner.
In the process of data acquisition experiments, continuous motion acquisition makes patients easily tired, and the acquired actions at the moment can be deformed or can not be completed according to the specified time. The data acquisition paradigm sets the execution time of each task to 15 seconds, taking into account the fatigue of the brain functional disorder patient in combination with the related work.
3. Data preprocessing method
In order to solve the influence of sensor errors existing in original kinematic signals and improve the analysis effect of the signals, the embodiment provides a preprocessing method of the kinematic signals.
Specifically, the present embodiment uses a fast fourier transform method for the filtering process of the original kinematic data. Next, the present embodiment uses a data enhancement algorithm to improve the generalization ability of the model. On this basis, the present embodiment uses a normalization algorithm to eliminate the impact of individual differences from patient to patient on data quality and model performance.
(1) Fast fourier transform-based filtering
In the process of a data acquisition experiment, the acceleration value at the initial moment is influenced by considering that the wearing position of each brain functional disease patient is slightly different. To eliminate the effect of this on the quality of the brain function disease patient kinematic data and the model performance, the raw kinematic data is mapped to the corresponding frequency domain using a fast fourier transform method. The zero frequency and high frequency amplitude of the data are then zeroed out in the frequency domain. On the basis, the processed frequency domain data is inverted to time domain data through a fast Fourier transform function, so that the filtering effect is achieved.
Specifically, for a finite time sequence of length N, dimension 2
The fast fourier transform formula of the sequence can be expressed as:
the inverse fast fourier transform can be expressed as
Of the formula (I)
Is a twiddle factor in the fast fourier transform.
(2) Model generalization capability improvement based on data enhancement algorithm
In order to make the limited data more valuable, this embodiment uses two data enhancement methods in turn.
In the embodiment, a sliding window method is used for expanding a data set, and then a Gaussian white noise adding mode is used for enhancing the generalization capability of the model.
In this example, the frequency of action for collecting patients with brain function disorder was high, and the time window length was set to 5 seconds according to the direction of a professional neurologist, so that 1000 data points were contained in each sample. Thus, the purpose of expanding the data set and improving the model identification precision can be achieved.
The gaussian white noise may be generated using a pseudo-random number generator. The amount of added noise is a configurable super parameter, and this embodiment recommends applying gaussian white noise randomly generated five times the original data to the enhanced data:
in the middle of
Representing kinematic data after enhanced data processing, < >>
Kinematic data representing patients with brain function diseases treated by filtering and sliding window +.>
Representing a normal distribution function with a mean value of 0 and a standard deviation of 0.1.
(3) Individual variance cancellation based on data normalization
In order to improve the recognition performance of the deep learning model, the data after data filtering and reinforcement pretreatment is normalized and then sent into the neural network model for training.
In the middle of
And->
Kinematic data before and after normalization, respectively.
4. Upper limb bradykinesia quantitative grading evaluation model based on time convolution network
Aiming at the evaluation of the upper limb bradykinesia degree of the brain functional disease patient, the embodiment designs a quantitative evaluation model of the upper limb bradykinesia degree of the brain functional disease based on a time convolution network. The model can objectively and accurately realize four level classifications of the severity of the upper limb movement delay of the brain function disease patient. As shown in fig. 2, the evaluation model provided in this embodiment includes an input layer, a plurality of residual blocks, a full connection layer and an output layer.
In the evaluation model provided in this embodiment, each residual block is composed of two weight normalization layers, two Dropout layers, two dilation-causal convolution layers, and two rectification activation function (ReLU) layers. In addition, the expansion factor of each residual block varies by an exponential power of 2. Specifically, the 1 st residual block expansion factor is 1, and the 6 th residual block expansion factor becomes 32.
Fig. 5 is a schematic diagram of a specific structure of an evaluation model provided by the present invention. As shown in fig. 5, the input data of the model is from the kinematic data of the brain functional disease patient. Preferably, the three-axis acceleration is synthesized as a single axis to reduce the data dimension, and the y-axis angular velocity with the strongest correlation with the motion process is selected as the input, so the dimension of the input data is 2. The time step selected after downsampling the time series is 200, and the form of the input data of the determined model is 200×1×2.
In the method, in the process of the invention,
，/>
，/>
respectively represent +.>
，/>
，/>
Kinematic data on three axes, +.>
Data representing the entire accelerometer after integration.
Inside the first residual block, the optimal convolution kernel size is 7×1, and the number of convolution filters is 128, so the depth of the feature map after the first layer convolution becomes 128.
In order to enable the time convolution network model to acquire a time sequence with any length, after the time convolution network model passes through the first residual block, 1×1 convolution is applied between the input sequence and the second residual block, so that the data output length and the input length of the residual block are ensured to be consistent, and the information of the over time is also kept, thereby accelerating the convergence process of the model and improving the accuracy of model evaluation.
In order to avoid the problems of gradient disappearance, gradient explosion and the like of the model, the present embodiment uses residual connection in the interior of the remaining 5 residual blocks. Preferably, the size of each convolution kernel is set to 7×1, and the number of convolution filters is set to 128. The main difference between the different residual blocks is that as the model deepens, the expansion factor d of each residual block is 1,2,4,8,16,32, respectively. The model layer number deepens and the expansion factors are added, so that the time convolution network model can learn more features in the input time sequence, the recognition accuracy of the model is improved, and meanwhile, in order to avoid the phenomenon of overfitting in the model training process, a Dropout layer with the super parameter of 0.3 is applied to each layer of convolution.
In a time convolution network architecture, the fully-connected layer flattens the data into a one-dimensional signal. The number of neurons in the output layer was 4, representing four degrees of bradykinesia (normal, mild, moderate and severe) in patients with brain function disease, respectively. The Softmax nonlinear function is applied in a model at an output layer as a classifier, the working process can be understood as that given input obtains probability distribution of various categories after passing through the Softmax function, and then the category with the highest probability is used as a prediction result according to the maximum likelihood estimation principle:
in the method, in the process of the invention,
probability of being a predicted category, +.>
Is the model's score for this category, N represents the number of categories.
The super parameters of the model provided in this embodiment include the number of residual block layers, the number of convolution kernels, the size of the convolution kernels, and Dropout factors. According to the control variable method, the optimal super-parameters are designed as follows: 6 residual blocks, 128 convolution kernels, a size of 1x7 convolution kernels, a Dropout factor of 0.3.
The embodiment combines the wearable equipment and the deep learning algorithm to realize the objective and quantitative evaluation of the lightweight portable brain function disease dyskinesia, and can realize the high-efficiency self-monitoring management of the brain function disease patients.
The time convolution network-based bradykinesia quantization step assessment device provided by the invention is described below, and the time convolution network-based bradykinesia quantization step assessment device described below and the time convolution network-based bradykinesia quantization step assessment method described above can be referred to correspondingly.
Fig. 6 is a schematic diagram of a device for quantitatively and hierarchically estimating bradykinesia based on a time convolution network according to the present invention, as shown in fig. 6, the device comprises:
an acquisition data module 600, configured to acquire kinematic data of a target object;
an output level module 610, configured to obtain a bradykinesia level of the target object based on the kinematic data and the trained bradykinesia quantitative assessment model;
wherein the bradykinesia quantitative assessment model is trained from a time convolution network from a sample kinematic data set with bradykinesia class labels.
Optionally, the bradykinesia quantization assessment model comprises an input layer, a residual error network, a full connection layer and an output layer, wherein the input layer, the residual error network, the full connection layer and the output layer are sequentially connected;
the residual network comprises a plurality of residual blocks which are connected in sequence, and any two adjacent residual blocks are connected through a convolution layer of 1 multiplied by 1.
Optionally, each residual block includes two convolution units, and each convolution unit has a structure that an expansion causal convolution layer, a weight normalization layer, a linear rectification function ReLU layer and a Dropout layer are sequentially connected.
Optionally, in the residual network, the expansion factor of the residual block connected with the input layer is 1, and the expansion factor of any residual block is 2 times of the expansion factor of the previous residual block.
Optionally, deriving a bradykinesia level of the target object based on the kinematic data and the trained bradykinesia quantitative assessment model comprises:
preprocessing the kinematic data, and inputting the preprocessed kinematic data into a trained quantitative evaluation model of the bradykinesia to obtain the bradykinesia grade of a target object;
the preprocessing comprises filtering processing and/or normalization processing;
wherein the filtering process includes:
obtaining frequency domain data corresponding to the time domain data by fast Fourier transform of the time domain data of the kinematic data, and carrying out fast Fourier inverse transform after zeroing amplitude corresponding to zero frequency and amplitude corresponding to frequency higher than a preset frequency threshold in the frequency domain data;
the normalization process is performed according to the following formula:
in the method, in the process of the invention,
kinematic data before normalization processing; />
Is the kinematic data after normalization treatment. />
Optionally, the sample kinematic data set is obtained according to the following manner:
acquiring a sample original kinematic data set;
performing one or more of filtering processing, data enhancement processing and normalization processing on the original kinematic data set of the sample to obtain the kinematic data set of the sample;
wherein, the data enhancement processing includes:
expanding the original kinematic data set of the sample by using a sliding window method; and/or the number of the groups of groups,
gaussian white noise is added to the original kinematic data set of the sample.
Optionally, the kinematic data of the target object includes one or more of:
kinematic data when the target object performs a rest task;
kinematic data of the target object when performing the stretching task;
kinematic data when the target object performs the rotation task.
Fig. 7 is a schematic diagram of an entity structure of an electronic device according to the present invention, as shown in fig. 7, the electronic device may include: processor 710, communication interface (Communications Interface) 720, memory 730, and communication bus 740, wherein processor 710, communication interface 720, memory 730 communicate with each other via communication bus 740. Processor 710 may invoke logic instructions in memory 730 to perform a method for quantitatively hierarchical assessment of bradykinesia based on a time convolution network, the method comprising:
acquiring kinematic data of a target object;
obtaining the bradykinesia grade of the target object based on the kinematic data and the trained bradykinesia quantitative evaluation model;
wherein the bradykinesia quantitative assessment model is trained from a time convolution network from a sample kinematic data set with bradykinesia class labels.
Further, the logic instructions in the memory 730 described above may be implemented in the form of software functional units and may be stored in a computer readable storage medium when sold or used as a stand alone product. Based on this understanding, the technical solution of the present invention may be embodied essentially or in a part contributing to the prior art or in a part of the technical solution, in the form of a software product stored in a storage medium, comprising several instructions for causing a computer device (which may be a personal computer, a server, a network device, etc.) to perform all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a random access Memory (RAM, random Access Memory), a magnetic disk, or an optical disk, or other various media capable of storing program codes.
In yet another aspect, the present invention also provides a non-transitory computer readable storage medium having stored thereon a computer program which, when executed by a processor, is implemented to perform the method of time-convolution network-based quantized fractional assessment of bradykinesia provided by the above methods, the method comprising:
acquiring kinematic data of a target object;
obtaining the bradykinesia grade of the target object based on the kinematic data and the trained bradykinesia quantitative evaluation model;
wherein the bradykinesia quantitative assessment model is trained from a time convolution network from a sample kinematic data set with bradykinesia class labels.
The apparatus embodiments described above are merely illustrative, wherein the elements illustrated as separate elements may or may not be physically separate, and the elements shown as elements may or may not be physical elements, may be located in one place, or may be distributed over a plurality of network elements. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art will understand and implement the present invention without undue burden.
From the above description of the embodiments, it will be apparent to those skilled in the art that the embodiments may be implemented by means of software plus necessary general hardware platforms, or of course may be implemented by means of hardware. Based on this understanding, the foregoing technical solution may be embodied essentially or in a part contributing to the prior art in the form of a software product, which may be stored in a computer readable storage medium, such as ROM/RAM, a magnetic disk, an optical disk, etc., including several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the method described in the respective embodiments or some parts of the embodiments.
Finally, it should be noted that: the above embodiments are only for illustrating the technical solution of the present invention, and are not limiting; although the invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims (10)
1. A method for quantitatively and hierarchically assessing bradykinesia based on a time convolution network, comprising:
acquiring kinematic data of a target object;
obtaining a bradykinesia grade of the target object based on the kinematic data and the trained bradykinesia quantitative evaluation model;
the bradykinesia quantitative evaluation model is obtained by training a time convolution network according to a sample kinematics data set with a bradykinesia grade label.
2. The method for quantitative hierarchical assessment of bradykinesia based on a time convolution network according to claim 1 wherein said quantitative assessment model of bradykinesia comprises an input layer, a residual network, a fully connected layer and an output layer, said input layer, said residual network, said fully connected layer and said output layer being connected in sequence;
the residual network comprises a plurality of residual blocks which are connected in sequence, and any two adjacent residual blocks are connected through a 1X 1 convolution layer.
3. The method for quantitative hierarchical estimation of bradykinesia based on time convolution network of claim 2, wherein each residual block comprises two convolution units, each convolution unit is composed of an expansion causal convolution layer, a weight normalization layer, a linear rectification function ReLU layer, a Dropout layer connected in turn.
4. A method of quantized motion retardation grading assessment based on a time convolution network according to claim 3, wherein in the residual network, the residual blocks connected to the input layer have a value of 1 and any one residual block has a value of 2 times the value of the previous residual block.
5. The method for quantitative hierarchical assessment of bradykinesia based on a time convolution network according to claim 1 wherein said deriving a grade of bradykinesia for said target object based on said kinematic data and a trained quantitative assessment model of bradykinesia comprises:
preprocessing the kinematic data, inputting the preprocessed kinematic data into a trained quantitative evaluation model of the bradykinesia to obtain the bradykinesia grade of the target object;
the preprocessing comprises filtering processing and/or normalization processing;
wherein the filtering process includes:
obtaining frequency domain data corresponding to the time domain data by fast Fourier transform, and performing inverse fast Fourier transform after setting the amplitude corresponding to zero frequency and the amplitude corresponding to frequency higher than a preset frequency threshold in the frequency domain data;
the normalization is performed according to the following formula:
in the method, in the process of the invention,
kinematic data before normalization processing; />
Is the kinematic data after normalization treatment.
6. The method of quantitative hierarchical assessment of bradykinesia based on a time convolution network according to claim 5 wherein said sample kinematic data set is obtained according to the following:
acquiring a sample original kinematic data set;
performing one or more of filtering processing, data enhancement processing and normalization processing on the original kinematic data set of the sample to obtain the kinematic data set of the sample;
wherein the data enhancement processing includes:
expanding the sample original kinematic data set using a sliding window method; and/or the number of the groups of groups,
gaussian white noise is added to the sample raw kinematic dataset.
7. The method of quantitative hierarchical assessment of bradykinesia based on a time convolution network according to claim 1 wherein the kinematic data of the target object comprises one or more of the following:
kinematic data when the target object performs a rest task;
kinematic data of the target object when performing the stretching task;
kinematic data when the target object performs the rotation task.
8. A time convolution network-based bradykinesia quantitative hierarchical assessment apparatus comprising:
the acquisition data module is used for acquiring the kinematic data of the target object;
the output grade module is used for obtaining the bradykinesia grade of the target object based on the kinematic data and the trained bradykinesia quantitative evaluation model;
the bradykinesia quantitative evaluation model is obtained by training a time convolution network according to a sample kinematics data set with a bradykinesia grade label.
9. An electronic device comprising a memory, a processor and a computer program stored on the memory and running on the processor, wherein the processor implements the method of quantitative hierarchical assessment of bradykinesia based on a time convolution network according to any one of claims 1 to 7 when executing the program.
10. A non-transitory computer readable storage medium having stored thereon a computer program, which when executed by a processor implements the method of quantized fractional assessment of bradykinesia based on a time convolution network according to any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310701314.3A CN116434908A (en) | 2023-06-14 | 2023-06-14 | Method and device for quantitatively and hierarchically evaluating bradykinesia based on time convolution network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310701314.3A CN116434908A (en) | 2023-06-14 | 2023-06-14 | Method and device for quantitatively and hierarchically evaluating bradykinesia based on time convolution network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116434908A true CN116434908A (en) | 2023-07-14 |
Family
ID=87091172
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310701314.3A Pending CN116434908A (en) | 2023-06-14 | 2023-06-14 | Method and device for quantitatively and hierarchically evaluating bradykinesia based on time convolution network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116434908A (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109480858A (en) * | 2018-12-29 | 2019-03-19 | 中国科学院合肥物质科学研究院 | It is a kind of for quantify detect disturbances in patients with Parkinson disease bradykinesia symptom wearable intelligence system and method |
| CN111938670A (en) * | 2020-06-30 | 2020-11-17 | 齐鲁工业大学 | Depression identification method and system based on human skeleton kinematics characteristic information |
| WO2022226439A1 (en) * | 2021-04-22 | 2022-10-27 | Verily Life Sciences Llc | Systems and methods for remote clinical exams and automated labeling of signal data |
| CN115336979A (en) * | 2021-09-03 | 2022-11-15 | 中国人民解放军总医院 | Multitask tremor automatic detection method based on wearable device and detection device thereof |
| CN116115193A (en) * | 2023-02-03 | 2023-05-16 | 电子科技大学 | Buchner stage rating prediction method integrating sensor array and TCN |
-
2023
- 2023-06-14 CN CN202310701314.3A patent/CN116434908A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109480858A (en) * | 2018-12-29 | 2019-03-19 | 中国科学院合肥物质科学研究院 | It is a kind of for quantify detect disturbances in patients with Parkinson disease bradykinesia symptom wearable intelligence system and method |
| CN111938670A (en) * | 2020-06-30 | 2020-11-17 | 齐鲁工业大学 | Depression identification method and system based on human skeleton kinematics characteristic information |
| WO2022226439A1 (en) * | 2021-04-22 | 2022-10-27 | Verily Life Sciences Llc | Systems and methods for remote clinical exams and automated labeling of signal data |
| CN115336979A (en) * | 2021-09-03 | 2022-11-15 | 中国人民解放军总医院 | Multitask tremor automatic detection method based on wearable device and detection device thereof |
| CN116115193A (en) * | 2023-02-03 | 2023-05-16 | 电子科技大学 | Buchner stage rating prediction method integrating sensor array and TCN |
Non-Patent Citations (3)
| Title |
|---|
| ROBBIN ROMIJNDERS ET AL: "A Deep Learning Approach for Gait Event Detection from a Single Shank-Worn IMU: Validation in Healthy and Neurological Cohorts", 《SENSORS》, vol. 22, no. 10, pages 3 - 11 * |
| 李古强 等: "《人体运动学》", 华中科技大学出版社, pages: 4 - 5 * |
| 李文涛: "基于面部特征和肢体动作多模态抑郁症识别研究", 《中国优秀硕士学位论文全文数据库医药卫生科技辑》, no. 9, pages 21 - 26 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Li et al. | An automatic user-adapted physical activity classification method using smartphones | |
| EP3836836B1 (en) | Real-time spike detection and identification | |
| CN116507276A (en) | Method and apparatus for machine learning to analyze musculoskeletal rehabilitation from images | |
| KR102363879B1 (en) | Method for predicting clinical functional assessment scale using feature values derived by upper limb movement of patients | |
| Senanayake et al. | A knowledge-based intelligent framework for anterior cruciate ligament rehabilitation monitoring | |
| Butt et al. | Fall detection using LSTM and transfer learning | |
| Setiawan et al. | Development of neuro-degenerative diseases’ gait classification algorithm using convolutional neural network and wavelet coherence spectrogram of gait synchronization | |
| Zhen et al. | Hybrid Deep‐Learning Framework Based on Gaussian Fusion of Multiple Spatiotemporal Networks for Walking Gait Phase Recognition | |
| Lotfi et al. | Classification of gait patterns in the time domain using accelerometer signal | |
| Li et al. | Neuralgait: Assessing brain health using your smartphone | |
| Aydin | Fuzzy Integral and Cuckoo Search Based Classifier Fusion for Human Action Recognition. | |
| Zeng et al. | Accelerometer-based gait recognition via deterministic learning | |
| Goh et al. | Multilayer perceptron neural network classification for human vertical ground reaction forces | |
| US20230329587A1 (en) | System And Method For Assessing Neuro Muscular Disorder By Generating Biomarkers From The Analysis Of Gait | |
| Ekambaram et al. | Artificial Intelligence (AI) Powered Precise Classification of Recuperation Exercises for Musculoskeletal Disorders | |
| CN116434908A (en) | Method and device for quantitatively and hierarchically evaluating bradykinesia based on time convolution network | |
| CN114171194B (en) | Quantitative assessment method, device, electronic device and medium for Parkinson multiple symptoms | |
| Vijayvargiya et al. | Implementation of machine learning algorithms for automated human gait activity recognition using sEMG signals | |
| Mostafa et al. | BioDeep: A Deep Learning System for IMU-based Human Biometrics Recognition. | |
| El-Ziaat et al. | Using Multi-Feature Fusion for Detecting Freezing of Gait Episodes in Patients with Parkinson's Disease | |
| Jahan et al. | Parkinson’s disease detection using resnet50 with transfer learning | |
| Tao et al. | Activity selection to distinguish healthy people from parkinson’s disease patients using i-da | |
| Mekruksavanich et al. | Refined lstm network for sensor-based human activity recognition in real world scenario | |
| Elwin et al. | Entropy Weighted and Kernalized Power K-Means Clustering Based Lesion Segmentation and Optimized Deep Learning for Diabetic Retinopathy Detection | |
| Johri et al. | Sensory Data Classification for Gait Assessment using Deep Neural Networks: A Comparative Study with SGD and Adam Optimizer |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20230714 |
|
| RJ01 | Rejection of invention patent application after publication |