CN116825335B - Method and apparatus for predictive model training of prognostic outcome of parkinson's patient gait - Google Patents
Method and apparatus for predictive model training of prognostic outcome of parkinson's patient gait Download PDFInfo
- Publication number
- CN116825335B CN116825335B CN202311103673.5A CN202311103673A CN116825335B CN 116825335 B CN116825335 B CN 116825335B CN 202311103673 A CN202311103673 A CN 202311103673A CN 116825335 B CN116825335 B CN 116825335B
- Authority
- CN
- China
- Prior art keywords
- gait
- patient
- loss function
- network model
- model
- 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
Links
- 230000005021 gait Effects 0.000 title claims abstract description 134
- 238000012549 training Methods 0.000 title claims abstract description 53
- 238000000034 method Methods 0.000 title claims abstract description 38
- 239000003814 drug Substances 0.000 claims abstract description 78
- 229940079593 drug Drugs 0.000 claims abstract description 77
- 230000000694 effects Effects 0.000 claims abstract description 63
- 208000027089 Parkinsonian disease Diseases 0.000 claims abstract description 27
- 206010034010 Parkinsonism Diseases 0.000 claims abstract description 27
- 238000004364 calculation method Methods 0.000 claims abstract description 9
- 230000006870 function Effects 0.000 claims description 82
- 230000004927 fusion Effects 0.000 claims description 31
- 238000010606 normalization Methods 0.000 claims description 15
- 230000002441 reversible effect Effects 0.000 claims description 8
- 239000013598 vector Substances 0.000 claims description 4
- 238000004393 prognosis Methods 0.000 abstract description 8
- 238000010586 diagram Methods 0.000 description 8
- 238000004891 communication Methods 0.000 description 6
- 238000011282 treatment Methods 0.000 description 6
- 206010056242 Parkinsonian gait Diseases 0.000 description 5
- -1 aminotricyclodecylamine Chemical compound 0.000 description 5
- 238000000605 extraction Methods 0.000 description 5
- 230000007659 motor function Effects 0.000 description 5
- 238000012545 processing Methods 0.000 description 5
- 238000013270 controlled release Methods 0.000 description 4
- 239000003136 dopamine receptor stimulating agent Substances 0.000 description 4
- 230000002093 peripheral effect Effects 0.000 description 4
- WTDRDQBEARUVNC-UHFFFAOYSA-N L-Dopa Natural products OC(=O)C(N)CC1=CC=C(O)C(O)=C1 WTDRDQBEARUVNC-UHFFFAOYSA-N 0.000 description 3
- 229940123685 Monoamine oxidase inhibitor Drugs 0.000 description 3
- 208000018737 Parkinson disease Diseases 0.000 description 3
- 230000003993 interaction Effects 0.000 description 3
- 229960004502 levodopa Drugs 0.000 description 3
- 239000002899 monoamine oxidase inhibitor Substances 0.000 description 3
- 230000008569 process Effects 0.000 description 3
- 238000012546 transfer Methods 0.000 description 3
- 230000005540 biological transmission Effects 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000002360 preparation method Methods 0.000 description 2
- 235000018553 tannin Nutrition 0.000 description 2
- 229920001864 tannin Polymers 0.000 description 2
- 239000001648 tannin Substances 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 230000001225 therapeutic effect Effects 0.000 description 2
- 230000000007 visual effect Effects 0.000 description 2
- 210000000707 wrist Anatomy 0.000 description 2
- 206010006100 Bradykinesia Diseases 0.000 description 1
- 206010017577 Gait disturbance Diseases 0.000 description 1
- 208000006083 Hypokinesia Diseases 0.000 description 1
- WTDRDQBEARUVNC-LURJTMIESA-N L-DOPA Chemical compound OC(=O)[C@@H](N)CC1=CC=C(O)C(O)=C1 WTDRDQBEARUVNC-LURJTMIESA-N 0.000 description 1
- 206010061533 Myotonia Diseases 0.000 description 1
- 206010071390 Resting tremor Diseases 0.000 description 1
- 206010044565 Tremor Diseases 0.000 description 1
- 210000003423 ankle Anatomy 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000001413 cellular effect Effects 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 229940052760 dopamine agonists Drugs 0.000 description 1
- 210000000245 forearm Anatomy 0.000 description 1
- 230000010365 information processing Effects 0.000 description 1
- 210000003127 knee Anatomy 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 208000015122 neurodegenerative disease Diseases 0.000 description 1
- 230000001144 postural effect Effects 0.000 description 1
- 230000007704 transition Effects 0.000 description 1
- 238000011269 treatment regimen Methods 0.000 description 1
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The application discloses a method and equipment for training a predictive model of the gait prognosis effect of a parkinsonism patient. The method comprises the following steps: acquiring pre-gait characteristics and medication information of a parkinsonism patient, and inputting the pre-gait characteristics and the medication information into the prediction model; based on the pre-gait characteristics and the medication information, carrying out weight calculation by using an auxiliary network model in the prediction model to obtain the weight corresponding to each Parkinson patient; based on the pre-gait characteristics and the medication information, extracting characteristics by using a backbone network in a prediction model and calculating an initial loss function; obtaining a total loss function by taking the weight as a coefficient of the initial loss function; and reversely training the main network model and the auxiliary network model by utilizing the total loss function so as to train a prediction model for predicting the prognosis effect of the gait of the Parkinson patient. By utilizing the scheme of the application, the stability and the precision of the prediction model can be improved, and an accurate prediction result of the prognosis effect of the gait of the patient can be obtained.
Description
Technical Field
The present application relates generally to the field of artificial intelligence. More particularly, the present application relates to a method, apparatus and computer readable storage medium for training a predictive model for predicting the prognostic effect of gait in parkinson's patients. Further, the present application relates to a method, apparatus and computer readable storage medium for predicting the prognostic effect of gait in parkinson's patients.
Background
Parkinson's Disease ("PD") is a common central nervous system degenerative Disease of middle-aged and elderly people, and its clinical manifestations mainly include resting tremor, bradykinesia, myotonia and postural gait disorder, and drug treatment is the most important treatment means for PD. However, the medicament for treating PD has a certain side effect, and long-term administration or unsuitable administration type may not have a therapeutic effect, but may cause harm to the body of the PD patient, so that the PD patient is painful. Therefore, knowing the therapeutic effect (or prognostic effect) in advance in order to stop or change the treatment regimen of PD patients in time, alleviating the suffering of PD patients is a urgent problem to be solved.
Along with the continuous development of deep learning, the problem is expected to be solved by predicting the treatment effect through a trained deep network model, but the existing model training is often directly trained on all training samples, so that the problem of unbalance of the training samples cannot be effectively treated. For example, by training directly on all training samples, the model is biased to learn in multiple samples, and less attention is paid to few samples, and important features in the few samples are ignored, so that the performance of the model is affected, and the stability and the accuracy of the model are poor.
In view of the foregoing, it is desirable to provide a solution for training a predictive model for predicting the prognostic effect of gait in parkinson's patients in order to improve the stability and accuracy of the predictive model so that accurate predictions of prognostic effects of gait in Guan Pajin sen patients can be obtained.
Disclosure of Invention
In order to solve at least one or more of the technical problems mentioned above, the present application proposes, in various aspects, a solution for training a predictive model for predicting the prognostic effect of gait in parkinson's patients.
In a first aspect, the present application provides a method for training a predictive model for predicting the prognostic effect of gait of a parkinson's patient, wherein the predictive model comprises a backbone network model and an auxiliary network model, and the method comprises: acquiring pre-gait characteristics and medication information of a parkinsonism patient, and inputting the pre-gait characteristics and the medication information into the prediction model; based on the pre-gait characteristics and the medication information, carrying out weight calculation by using an auxiliary network model in the prediction model to obtain the weight corresponding to each Parkinson patient; extracting features and calculating an initial loss function by using a backbone network in the prediction model based on the pre-gait features and the medication information; obtaining a total loss function by taking the weight as a coefficient of the initial loss function; and reverse training the backbone network model and the auxiliary network model using the total loss function to train a predictive model that predicts a prognostic outcome of parkinson's patient gait.
In one embodiment, wherein the pre-gait feature is characterized via a scale associated with gait of the parkinson's patient, and the scale comprises one or more of an H & Y scale, a TUG scale, or a Berg scale; the medication information includes at least a medication type and a medication dosage.
In another embodiment, before inputting the pre-gait feature and the medication information into the predictive model, further comprising: and performing feature fusion operation on the pre-gait feature and the medication information by using a feature fusion model so as to obtain fusion features.
In yet another embodiment, the method further comprises: and selecting a weight larger than a preset threshold as a coefficient of the initial loss function to obtain the total loss function.
In yet another embodiment, wherein the total loss function is obtained by: performing normalization operation on the weights to obtain weight normalization results; and taking the weight normalization result as a coefficient of the initial loss function to obtain the total loss function.
In yet another embodiment, wherein back training the backbone network model and the auxiliary network model with the total loss function comprises: fixing parameters of the auxiliary network model, and calculating a first gradient of the total loss function; and inversely updating the backbone network model according to a first gradient of the total loss function.
In yet another embodiment, wherein back training the backbone network model and the auxiliary network model with the total loss function further comprises: fixing parameters of the backbone network model, and calculating a second gradient of the total loss function; and inversely updating the auxiliary network model according to a second gradient of the total loss function.
In a second aspect, the present application provides a method for predicting the prognostic effect of gait in a parkinson's patient, comprising: acquiring pre-gait characteristics and medication information of a patient with parkinsonism to be predicted; and inputting the pre-gait feature and the medication information into a prediction model trained according to the method according to the plurality of embodiments in the first aspect to predict, so as to output a prediction result of predicting the prognosis effect of the gait of the parkinsonism patient.
In a third aspect, the present application provides an apparatus for training a predictive model for predicting the prognostic effect of gait in a parkinson's patient and for predicting the prognostic effect of gait in a parkinson's patient, comprising: a processor; and a memory in which program instructions for training a predictive model for predicting a prognostic effect of a gait of a parkinson's patient and for predicting a prognostic effect of a gait of a parkinson's patient are stored, which program instructions, when executed by the processor, cause the apparatus to carry out the plurality of embodiments of the first aspect and to carry out the embodiments of the second aspect.
In a fourth aspect, the present application provides a computer readable storage medium having stored thereon computer readable instructions for training a predictive model for predicting a prognostic effect of gait of a parkinson's patient and for predicting a prognostic effect of gait of a parkinson's patient, which computer readable instructions, when executed by one or more processors, implement the various embodiments of the foregoing first aspect and implement the embodiments of the foregoing second aspect.
By the scheme provided above for training the prediction model for predicting the prognostic effect of the gait of the parkinson patient, the embodiment of the application calculates the corresponding weight of each sample (i.e., each parkinson patient) based on the pre-gait characteristics and medication information of the parkinson patient by using the auxiliary network model, and uses the weight as the coefficient of the initial loss function calculated by the main network based on the pre-gait characteristics and medication information extraction characteristics, thereby training the prediction model by using the total loss function after the distribution of the coefficients. Based on the method, the importance of each sample can be adaptively adjusted, so that the prediction model pays attention to important features, and the performance and the accuracy of the prediction model are greatly improved. Furthermore, the embodiment of the application also helps the prediction model pay more attention to important features by selecting the weight higher than the threshold value as the coefficient of the initial loss function, and improves the prediction performance. Furthermore, the embodiment of the application also trains the main network or the auxiliary network model independently by fixing the parameters of the auxiliary network model or the main network, thereby reducing the training complexity and improving the training efficiency.
In addition, the embodiment of the application predicts the prognosis effect of the gait of the parkinsonism patient by using the prediction model which is completed by training, so that whether the gait of the parkinsonism patient is changed after intervention (such as drug treatment) can be known as early as possible and accurately, and the intervention scheme is stopped or changed in time for the parkinsonism patient which is unchanged, so that the prognosis effect of the gait of the parkinsonism patient is improved, and the pain of the parkinsonism patient is relieved.
Drawings
The above, as well as additional purposes, features, and advantages of exemplary embodiments of the present application will become readily apparent from the following detailed description when read in conjunction with the accompanying drawings. In the drawings, embodiments of the application are illustrated by way of example and not by way of limitation, and like reference numerals refer to similar or corresponding parts and in which:
FIG. 1 is an exemplary flow chart illustrating a method for training a predictive model for predicting the prognostic effect of a Parkinson patient's gait in accordance with an embodiment of the application;
FIG. 2 is an exemplary schematic diagram illustrating training of a predictive model for predicting the prognostic effect of gait in a Parkinson patient in accordance with an embodiment of the application;
FIG. 3 is yet another exemplary schematic diagram illustrating training of a predictive model for predicting the prognostic effect of gait of a Parkinson patient in accordance with an embodiment of the application;
FIG. 4 is an exemplary schematic diagram illustrating a method for predicting the prognostic effect of gait in a Parkinson patient, according to an embodiment of the application; and
FIG. 5 is an exemplary block diagram illustrating an apparatus for training a predictive model for predicting the prognostic effect of a Parkinson's patient's gait and for predicting the prognostic effect of a Parkinson's patient's gait in accordance with an embodiment of the application.
Detailed Description
The technical solutions of the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. It should be understood that the embodiments described in this specification are only some embodiments of the application provided for the purpose of facilitating a clear understanding of the solution and meeting legal requirements, and not all embodiments of the application may be implemented. All other embodiments, which can be made by those skilled in the art without making any inventive effort, are intended to be within the scope of the present application based on the embodiments disclosed herein.
FIG. 1 is an exemplary flow chart illustrating a method 100 for training a predictive model for predicting the prognostic effect of a parkinsonian gait in accordance with an embodiment of the application. As shown in fig. 1, at step S101, pre-gait characteristics and medication information of a parkinson patient are acquired and input into a predictive model. In one embodiment, the aforementioned pre-gait characteristics of the parkinsonism patient may be characterized via a scale associated with the parkinsonism patient's gait, and the scale may include, but is not limited to, one or more of an H & Y scale, a TUG scale, or a Berg scale. For example, the scale may also be the StrLength_fp (cm) scale. The aforementioned administration information may include, but is not limited to, a drug type and a drug dosage, wherein the drug type may be, for example, various drug types such as dopamine agonist, a tannin controlled release tablet, a monoamine oxidase inhibitor, aminotricyclodecylamine, and a levodopa compound preparation, and the aforementioned drug dosage includes, for example, daily dosage and the like.
It will be appreciated that the H & Y scale is a scale for assessing motor function in PD patients and includes six assessment items, namely, flexion, extension of the upper limb forearm, dorsiflexion of the wrist, palmar flexion of the wrist, extension of the lower limb knee, and dorsiflexion of the ankle, with each assessment item ranging from 0-5 points, a total score of 30 points, and the higher the score, the better the motor function characterizing the PD patient; otherwise, the worse the motor function of the PD patient is characterized. The TUG scale is a scale for evaluating the walking and balance ability of a PD patient, which tests the time required for the PD patient to stand up from a chair, walk a distance, turn back to sit down on the chair, to characterize the walking and balance ability of the PD patient based on the length of time. The Berg scale is a scale for assessing the balance of a PD patient and includes a plurality of assessment items such as the balance, gait, sitting posture transition, etc. of the PD patient, each with a different score, the higher the score is, the better the balance of the PD patient is. The StrLength_fp (cm) scale is a scale used to evaluate the gait of PD patients, whose gait problems are evaluated by measuring the distance the PD patient has travelled at different speeds, the lower the StrLength_fp (cm) value, the more severe the gait problems characterizing the PD patient. In an implementation scenario, the pre-gait characteristics of parkinson's patients may be characterized in particular by scores of the various scales described above.
In one implementation scenario, before the pre-gait feature and the medication information are input into the prediction model, a feature fusion model may be used to perform a feature fusion operation on the pre-gait feature and the medication information to obtain a fusion feature, and then the fusion feature is input into the prediction model. The fusion characteristic can capture the complex relation between the pre-gait characteristic and the medication information. In some embodiments, the pre-gait features and the medication information may also be normalized prior to performing the feature fusion operation using the feature fusion model, e.g., the pre-gait features and the medication information may be "one-hot" encoded to process the pre-gait features and the medication information into a form suitable for the type of feature fusion model input. In some implementations, the aforementioned feature fusion model may be, for example, a fully connected network model.
As an example, for a drug type, it may be treated, for example, [10000], [01000], [00100], [00010] and [00001], to correspond to the indicated treatment of parkinson patients with dopamine agonists, a controlled release tablet of rest, monoamine oxidase inhibitors, aminotricyclodecylamine and levodopa. For pre-gait characteristics, characterized by the H & Y scale, the TUG scale and the Berg scale, it can also handle [101], correspondingly indicating good motor function, poor walking and balance ability and good balance force for parkinson patients. For each parkinsonian patient, the pre-gait feature and the medication information are typically combined and represented, for example, as [101] and [10000] for the pre-gait feature, which is combined and represented as [10110000]. Further, the data of each Parkinson patient after the normalization operation is input into a feature fusion model to perform feature fusion, and then the fusion features are input into a prediction model.
In one embodiment, the predictive model may include a backbone network model and an auxiliary network model. At step S102, based on the pre-gait characteristics and medication information, a weight calculation is performed using an auxiliary network model in the predictive model to obtain a weight corresponding to each parkinson patient. That is, the pre-gait characteristics and medication information of each parkinson patient are used as inputs of the auxiliary network model, weight calculation is performed via the auxiliary network model, and the weight corresponding to each parkinson patient (i.e., each sample) is output. At step S103, features are extracted and an initial loss function is calculated using the backbone network in the predictive model, also based on the pre-gait features and the medication information. That is, with the pre-gait characteristics and medication information of each parkinson patient as inputs to the backbone network, the prediction results are output after the feature extraction operation is performed via the backbone network, and then the initial loss function can be calculated from the prediction results and the true values. In some embodiments, the backbone network model and the auxiliary network model in the foregoing predictive model may also be, for example, fully connected networks.
Next, at step S104, a total loss function is obtained using the weights as coefficients of the initial loss function. In one implementation scenario, the weights of all samples may be selected as coefficients of the initial loss function to obtain the total loss function; weights greater than a preset threshold may also be selected as coefficients of the initial loss function to obtain a total loss function to focus more on learning of important samples to further improve the performance of the predictive model. In one embodiment, the total loss function may be obtained by performing a normalization operation on the weights to obtain a weight normalization result, and further using the weight normalization result as a coefficient of the initial loss function to obtain the total loss function. That is, the processing is performed by normalizing the selected weights to be the sum of them to 1 to convert the weights into probabilities.
Based on the total loss function obtained above, at step S105, the trunk network model and the auxiliary network model are reversely trained with the total loss function to train a predictive model for predicting the prognostic effect of the gait of the parkinson patient. That is, the parameters in the backbone network model and the auxiliary network model are reversely adjusted based on the total loss function so that the prediction result of the prediction model approaches to the true value to complete the training of the prediction model. Because of the high complexity of adjusting parameters in both the backbone network model and the auxiliary network model, embodiments of the present application further provide for fixing one of the networks while adjusting the other network. For example, in one implementation scenario, the first gradient of the total loss function may be calculated by fixing parameters of the auxiliary network model, and then updating the backbone network model inversely according to the first gradient of the total loss function. In another implementation scenario, the second gradient of the total loss function may also be calculated by fixing parameters of the backbone network model, and the auxiliary network model is updated reversely according to the second gradient of the total loss function.
As can be seen from the above description, the embodiment of the present application performs weight calculation by inputting the pre-gait characteristics and medication information of the parkinson's patient to the auxiliary network model in the prediction model, performs feature extraction and calculates an initial loss function by inputting the pre-gait characteristics and medication information of the parkinson's patient to the backbone network in the prediction model, and then selects weights of all samples or weights higher than a threshold value as coefficients of the initial loss function. Therefore, the importance of each sample can be adaptively adjusted by the prediction model, the intrinsic law of important features is more focused and adaptively learned, and the performance and the accuracy of the prediction model are greatly improved. In a scene with only a small number of training samples, the scheme of the embodiment of the application can enable the model to pay attention to important features in the small number of samples, so that the performance of the prediction model can be improved under the condition of only the small number of samples. Furthermore, the embodiment of the application also trains the main network or the auxiliary network model independently by fixing the parameters of the auxiliary network model or the main network, so as to reduce the training complexity and improve the training efficiency.
FIG. 2 is an exemplary schematic diagram illustrating training of a predictive model for predicting the prognostic effect of gait in a Parkinson patient in accordance with an embodiment of the application. It should be appreciated that FIG. 2 is a specific embodiment of the method 100 of FIG. 1 described above, and thus the description of FIG. 1 described above applies equally to FIG. 2.
As shown in fig. 2, the predictive model 201 of an embodiment of the application may include a backbone network model 202 and an auxiliary network model 203. In one implementation scenario, the backbone network model 202 and the auxiliary network model 203 may be, for example, fully connected networks. In this scenario, the pre-gait feature 204 and the medication information 205 of the parkinsonism patient are first input to the auxiliary network model 203 for weight calculation, the corresponding weight 206 for each parkinsonism patient is obtained, then the pre-gait feature 204 and the medication information 205 of the parkinsonism patient are input to the backbone network model 202 for feature extraction, and the initial loss function 207 is calculated. As can be seen from the foregoing, before the pre-gait feature 204 and the medication information 205 of the parkinson's patient are input to the predictive model 201, the feature fusion model may perform a feature fusion operation on the pre-gait feature 204 and the medication information 205 to obtain a fusion feature, which is then input to the predictive model 201.
In some embodiments, the foregoing pre-gait characteristics may be characterized via scales associated with the gait of the parkinson's patient (e.g., H & Y scale, TUG scale, or Berg scale), particularly by scoring scores of the scales to characterize, for example, the motor function, walking and balance ability, or the balance of the parkinson's patient. The aforementioned medication information may be, for example, the type of drug (e.g., dopamine agonist, a controlled-release tablet of rest, etc.) and the dosage of the drug (e.g., daily dose, etc.). In one implementation scenario, prior to performing the feature fusion operation using the feature fusion model, the pre-gait features and medication information may also be normalized (e.g., in one-hot encoding) to process the pre-gait features and medication information into a form suitable for the type of feature fusion model input (e.g., a feature vector form such as [10110000 ]).
In one exemplary scenario, assuming that the pre-gait feature 204 and medication information 205 of a parkinson patient are denoted as feature vectors h, the auxiliary network model 203 is denoted as g, the weights calculated via the auxiliary network model 203 may be expressed as w=g (h). Assuming that the backbone network model 202 is denoted as m, and its output prediction and true values are denoted as y_hat (y_hat=m (h)) and y, respectively, the initial Loss function Loss (y, y_hat) can be calculated. In some embodiments, the initial Loss function Loss (y, y_hat) may be, for example, a mean square error Loss or a cross entropy Loss. Further, by taking the weight 206 of each parkinsonism patient as a coefficient of the initial loss function 207, the total loss function 208 can be obtained. As an example, the total Loss function l=w×loss (y, y_hat). The aforementioned backbone network model 202 and auxiliary network model 203 are then trained in reverse using the total loss function 208 to achieve training of a predictive model that predicts the prognostic effect of parkinsonian gait.
It may be understood that the foregoing weight w=g (h) represents a weight for which the normalization operation is not performed, and the weight normalization result obtained by normalizing the weight w=g (h) may be expressed as w_normalized=w/sum (w_1, w_2,) and w_n, where the value of w is w_1, w_2,) and w_n, N represents the number of samples, and sum is equal to the sum) Represents the sum, w_1, w_2,..w_n represents the weight corresponding to each sample. In this scenario, the total Loss function l=w_normalized×loss (y, y_hat). That is, the weight normalization result is taken as a coefficient of the initial Loss function Loss (y, y_hat). In some embodiments, a weight wt greater than the threshold t may also be selected, followed by the normalization operation described above to determine a weight normalization result, e.g., w_normalized = wt/sum (wt), where wt represents a weight where w is greater than t.
After the total loss function is obtained, the main network model and the auxiliary network model can be trained reversely by using the total loss function. In one embodiment, the first gradient of the total loss function may be calculated by fixing parameters of the auxiliary network model. Assuming that the first gradient of the total loss function is denoted as gradient_m, gradient_m=dl/dy_hat×dh/dm, the backbone network model may be updated inversely according to the first gradient of the total loss function denoted as gradient_m. In another embodiment, the second gradient of the total loss function may also be calculated by fixing parameters of the backbone network model. Assuming that the second gradient of the total loss function is denoted gradient_g, gradient_g=dl/dw×dh/dg, the auxiliary network model may be updated inversely according to the second gradient of the total loss function. Therefore, the training complexity can be reduced, and the training efficiency is improved.
FIG. 3 is yet another exemplary schematic diagram illustrating training of a predictive model for predicting the prognostic effect of gait in a Parkinson patient in accordance with an embodiment of the application. It should be appreciated that FIG. 3 is yet another embodiment of the method 100 of FIG. 1 described above, and thus the description of FIG. 1 described above applies equally to FIG. 3.
As shown in fig. 3, the predictive model 201 of an embodiment of the application may include a backbone network model 202 and an auxiliary network model 203. In one implementation scenario, the backbone network model 202 and the auxiliary network model 203 may be, for example, fully connected networks. In this scenario, feature fusion model 301 is first used to perform feature fusion on pre-gait features 204 and medication information 205 of a parkinson patient to obtain fusion features, which are then input into predictive model 201. As previously described, prior to performing the feature fusion operation using the feature fusion model 301, the pre-gait features and medication information may also be normalized (e.g., in one-hot encoding) to process the pre-gait features and medication information into a form suitable for the type of feature fusion model input (e.g., a feature vector form such as [10110000 ]).
As further shown, the fused features output by the feature fusion model 301 are input to the auxiliary network model 203 for weight calculation to obtain weights 206 corresponding to each parkinson's patient, and then the pre-gait features 204 and medication information 205 of the parkinson's patient are input to the main network model 202 for feature extraction, and an initial loss function 207 is calculated. For more details on the specific foregoing weight and initial loss function calculation, reference may be made to the descriptions of fig. 1 and fig. 2, and the present application is not repeated here.
Further, by taking the weight 206 of each parkinsonism patient as a coefficient of the initial Loss function 207, a total Loss function 208 may be obtained, which may be expressed as l=w×loss (y, y_hat), where w represents the weight and Loss (y, y_hat) represents the initial Loss function, for example. The aforementioned backbone network model 202 and auxiliary network model 203 are then trained in reverse using the total loss function 208 to achieve training of a predictive model that predicts the prognostic effect of parkinsonian gait. Specifically, the backbone network model and the auxiliary network model may be trained in reverse at the same time according to the total loss function 208, or one of the backbone network model and the auxiliary network model may be fixed, and the gradient of the corresponding total loss function may be calculated to update the other network in reverse. For more details on the gradient of the total loss function, reference is made to the description of fig. 1 and 2 above, and the application is not repeated here.
FIG. 4 is an exemplary schematic diagram illustrating a method 400 for predicting the prognostic effect of gait in a Parkinson patient, according to an embodiment of the application. As shown in fig. 4, at step S401, pre-gait characteristics and medication information of a parkinson' S disease patient to be predicted are acquired. In one embodiment, the aforementioned pre-gait characteristics of the parkinsonism patient may be characterized via a scale associated with the parkinsonism patient's gait, and the scale may include, but is not limited to, one or more of an H & Y scale, a TUG scale, or a Berg scale. For example, the scale may also be the StrLength_fp (cm) scale. The aforementioned administration information may include, but is not limited to, a drug type and a drug dosage, wherein the drug type may be, for example, a dopamine agonist, a tannin controlled release tablet, a monoamine oxidase inhibitor, aminotricyclodecylamine, and a levodopa compound preparation, and the like, and the aforementioned drug dosage may include, for example, a daily dose, and the like. Next, at step S402, the pre-gait feature and the medication information are input into a prediction model that is trained to perform prediction, so as to output a prediction result of predicting the prognostic effect of the gait of the parkinson patient. That is, the prediction model trained by the training method according to the embodiment of the present application predicts based on the pre-gait characteristics and medication information of the parkinson's disease patient, so as to output the prediction result of the prognostic effect.
It should be understood that, the prediction model of the embodiment of the present application outputs a probability value regarding the prognostic effect of the gait of the parkinson patient, and by comparing the probability value with the preset threshold, the prognostic effect of the gait of the parkinson patient is considered to be good when the probability value is greater than the preset threshold. That is, the gait of the parkinsonism patient changes after the intervention, and the intervention regimen can continue. And when the probability value is smaller than a preset threshold value, the prognosis effect of the gait of the parkinsonism patient is considered to be poor. That is, the gait of the parkinsonism patient is unchanged after intervention, and the intervention scheme can be stopped or replaced in time. Therefore, the embodiment of the application can know whether the gait of the parkinsonism patient is changed or not as soon as possible and accurately after intervention (such as drug treatment), so as to improve the prognosis effect of the gait of the parkinsonism patient as soon as possible and relieve the pain of the parkinsonism patient.
Fig. 5 is an exemplary block diagram illustrating an apparatus 500 for training a predictive model for predicting the prognostic effect of gait of a parkinson's patient and for predicting the prognostic effect of gait of a parkinson's patient in accordance with an embodiment of the application. It is to be appreciated that the device implementing aspects of the present application may be a single device (e.g., a computing device) or a multi-function device including various peripheral devices.
As shown in fig. 5, the apparatus of the present application may include a central processing unit or central processing unit ("CPU") 511, which may be a general purpose CPU, a special purpose CPU, or other information processing and program running execution unit. Further, device 500 may also include a mass memory 512 and a read only memory ("ROM") 513, wherein mass memory 512 may be configured to store various types of data, including various and pre-gait characteristics and medication information, algorithm data, intermediate results and various programs required to operate device 500. ROM 513 may be configured to store data and instructions necessary to power-on self-test for device 500, initialization of functional modules in the system, drivers for basic input/output of the system, and boot the operating system.
Optionally, the device 500 may also include other hardware platforms or components, such as a tensor processing unit ("TPU") 514, a graphics processing unit ("GPU") 515, a field programmable gate array ("FPGA") 516, and a machine learning unit ("MLU") 517, as shown. It will be appreciated that while various hardware platforms or components are shown in device 500, this is by way of example only and not limitation, and that one of skill in the art may add or remove corresponding hardware as desired. For example, device 500 may include only a CPU, associated memory device, and interface device to implement the methods of the present application for training a predictive model for predicting the prognostic effect of a parkinsonian gait and methods for predicting the prognostic effect of a parkinsonian gait.
In some embodiments, to facilitate the transfer and interaction of data with external networks, the device 500 of the present application further comprises a communication interface 518, whereby the device can be connected to a local area network/wireless local area network ("LAN/WLAN") 505 via the communication interface 518, and further to a local server 506 or to the Internet ("Internet") 507 via the LAN/WLAN. Alternatively or additionally, the device 500 of the present application may also be directly connected to the internet or cellular network via the communication interface 518 based on wireless communication technology, such as wireless communication technology based on generation 3 ("3G"), generation 4 ("4G"), or generation 5 ("5G"). In some application scenarios, the device 500 of the present application may also access the server 508 and database 509 of the external network as needed to obtain various known algorithms, data, and modules, and may store various data remotely, such as various types of data or instructions for presenting, for example, pre-gait characteristics and medication information, etc.
The peripheral devices of the apparatus 500 may include a display device 502, an input device 503, and a data transmission interface 504. In one embodiment, display device 502 may, for example, include one or more speakers and/or one or more visual displays configured for voice prompts and/or visual display of the predictive model of the present application for training the prognostic effect of predicting gait of a parkinson's patient and the prognostic effect of predicting gait of a parkinson's patient. The input device 503 may include other input buttons or controls, such as a keyboard, mouse, microphone, gesture-capturing camera, etc., configured to receive input of audio data and/or user instructions. The data transfer interface 504 may include, for example, a serial interface, a parallel interface, or a universal serial bus interface ("USB"), a small computer system interface ("SCSI"), serial ATA, fireWire ("FireWire"), PCI Express, and high definition multimedia interface ("HDMI"), etc., configured for data transfer and interaction with other devices or systems. In accordance with aspects of the application, the data transmission interface 504 may receive the pre-gait characteristics and medication information from the medical database collection and transmit data or results including the pre-gait characteristics and medication information or various other types to the device 500.
The above-described CPU 511, mass memory 512, ROM 513, TPU 514, GPU 515, FPGA 516, MLU 517 and communication interface 518 of the device 500 of the present application may be interconnected by a bus 519 and data interaction with peripheral devices may be achieved by the bus. In one embodiment, CPU 511 may control other hardware components in device 500 and its peripherals through this bus 519.
The apparatus for training a predictive model for predicting the prognostic effect of gait of a parkinson's patient and for predicting the prognostic effect of gait of a parkinson's patient, which may be used to implement the present application, is described above in connection with fig. 5. It is to be understood that the device structure or architecture herein is merely exemplary and that the implementation and implementation entities of the present application are not limited thereto, but that changes may be made without departing from the spirit of the present application.
Those skilled in the art will also appreciate from the foregoing description, taken in conjunction with the accompanying drawings, that embodiments of the present application may also be implemented in software programs. The present application thus also provides a computer readable storage medium having stored thereon computer readable instructions for training a predictive model for predicting the prognostic effect of a gait of a parkinson's patient and for predicting the prognostic effect of a gait of a parkinson's patient, which when executed by one or more processors, may be used to implement the method of the present application for training a predictive model for predicting the prognostic effect of a gait of a parkinson's patient and the method for predicting the prognostic effect of a gait of a parkinson's patient as described in connection with figures 1 and 4.
It should be noted that although the operations of the method of the present application are depicted in the drawings in a particular order, this does not require or imply that the operations must be performed in that particular order or that all of the illustrated operations be performed in order to achieve desirable results. Rather, the steps depicted in the flowcharts may change the order of execution. Additionally or alternatively, certain steps may be omitted, multiple steps combined into one step to perform, and/or one step decomposed into multiple steps to perform.
It should be understood that when the terms "first," "second," "third," and "fourth," etc. are used in the claims, the specification and the drawings of the present application, they are used merely to distinguish between different objects, and not to describe a particular order. The terms "comprises" and "comprising" when used in the specification and claims of the present application are taken to specify the presence of stated features, integers, steps, operations, elements, and/or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and/or groups thereof.
It is also to be understood that the terminology used in the description of the application herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. As used in the specification and claims, the singular forms "a", "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It should be further understood that the term "and/or" as used in the present specification and claims refers to any and all possible combinations of one or more of the associated listed items, and includes such combinations.
Although the embodiments of the present application are described above, the descriptions are merely examples for facilitating understanding of the present application, and are not intended to limit the scope and application of the present application. Any person skilled in the art can make any modification and variation in form and detail without departing from the spirit and scope of the present disclosure, but the scope of the present disclosure is defined by the appended claims.
Claims (7)
1. A method for training a predictive model for predicting the prognostic outcome of gait in a parkinson's patient, said method comprising:
acquiring pre-gait characteristics and medication information of a parkinsonism patient, and inputting the pre-gait characteristics and the medication information into a prediction model, wherein the prediction model comprises a main network model and an auxiliary network model, and the main network model and the auxiliary network model are fully connected networks;
based on the pre-gait characteristics and the medication information, carrying out weight calculation by using an auxiliary network model in the prediction model to obtain the weight corresponding to each Parkinson patient;
based on the pre-gait characteristics and the medication information, extracting characteristics by using a backbone network in the prediction model, outputting a prediction result, and calculating an initial loss function according to the prediction result and a true value;
selecting weights larger than a preset threshold as coefficients of the initial loss function to obtain a total loss function;
training the backbone network model and the auxiliary network model in a reverse direction using the total loss function to train a predictive model that predicts a prognostic effect of parkinsonian patient gait, wherein the training the backbone network model and the auxiliary network model in a reverse direction using the total loss function comprises:
fixing parameters of the auxiliary network model, and calculating a first gradient of the total loss function; and
the backbone network model is updated inversely according to the first gradient of the total loss function,
wherein said calculating a first gradient of said total loss function comprises:
gradientm=dl/dy_hat×dh/dm, where gradientm represents a first gradient of the total loss function, L represents the total loss function, y_hat represents a prediction result output by the backbone network model, h represents a feature vector formed by the pre-gait feature and the medication information, and m represents the backbone network model;
the training the backbone network model and the auxiliary network model in reverse using the total loss function further comprises:
fixing parameters of the backbone network model, and calculating a second gradient of the total loss function; and
reversely updating the auxiliary network model according to a second gradient of the total loss function;
wherein said calculating a second gradient of said total loss function comprises:
gradent_g = dL/dw x dh/dg, where gradent_g represents the second gradient of the total loss function, w represents weights calculated via the auxiliary network model, and g represents the auxiliary network model.
2. The method of claim 1, wherein the pre-gait feature is characterized via a scale associated with gait of the parkinson's patient, and the scale comprises one or more of an H & Y scale, a TUG scale, or a Berg scale; the medication information includes at least a medication type and a medication dosage.
3. The method of claim 2, further comprising, prior to entering the pre-gait feature and the medication information into the predictive model:
and performing feature fusion operation on the pre-gait feature and the medication information by using a feature fusion model so as to obtain fusion features.
4. The method according to claim 1, characterized in that the total loss function is obtained by:
performing normalization operation on the weights to obtain weight normalization results; and
and taking the weight normalization result as a coefficient of the initial loss function to obtain the total loss function.
5. A method for predicting the prognostic effect of gait in a parkinson's patient, comprising:
acquiring pre-gait characteristics and medication information of a patient with parkinsonism to be predicted; and
inputting the pre-gait feature and the medication information into a prediction model trained according to the method of any one of claims 1-4 for prediction to output a prediction result of the prognostic effect of predicting the gait of the parkinson patient.
6. An apparatus for training a predictive model for predicting the prognostic effect of gait in a parkinson's patient and for predicting the prognostic effect of gait in a parkinson's patient, comprising:
a processor; and
a memory in which program instructions for training a predictive model for predicting a prognostic effect of gait of a parkinson's patient and for predicting a prognostic effect of gait of a parkinson's patient are stored, which program instructions, when executed by the processor, cause the apparatus to carry out the method according to any one of claims 1-4 or to carry out the method according to claim 5.
7. A computer readable storage medium having stored thereon computer readable instructions for training a predictive model for predicting a prognostic effect of a gait of a parkinson's patient and for predicting a prognostic effect of a gait of a parkinson's patient, which computer readable instructions, when executed by one or more processors, implement the method of any of claims 1-4 or the method of claim 5.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311103673.5A CN116825335B (en) | 2023-08-30 | 2023-08-30 | Method and apparatus for predictive model training of prognostic outcome of parkinson's patient gait |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311103673.5A CN116825335B (en) | 2023-08-30 | 2023-08-30 | Method and apparatus for predictive model training of prognostic outcome of parkinson's patient gait |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN116825335A CN116825335A (en) | 2023-09-29 |
| CN116825335B true CN116825335B (en) | 2023-12-08 |
Family
ID=88118859
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311103673.5A Active CN116825335B (en) | 2023-08-30 | 2023-08-30 | Method and apparatus for predictive model training of prognostic outcome of parkinson's patient gait |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116825335B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117789988B (en) * | 2024-02-27 | 2024-06-11 | 首都医科大学附属北京友谊医院 | Method for training predictive model for predicting parkinsonism and related products |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113012815A (en) * | 2021-04-06 | 2021-06-22 | 西北工业大学 | Parkinson health risk assessment method based on multi-modal data |
| CN113963798A (en) * | 2021-10-15 | 2022-01-21 | 华中科技大学同济医学院附属协和医院 | Parkinson's disease motor symptom evaluation method |
| CN115762757A (en) * | 2022-11-09 | 2023-03-07 | 电子科技大学 | Prediction system for improving postoperative motion of Parkinson's disease based on individual characteristics |
| CN115762720A (en) * | 2022-11-17 | 2023-03-07 | 吾征智能技术(北京)有限公司 | System, method, electronic device and storage medium for assessing Parkinson's disease efficacy |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8150629B2 (en) * | 2005-11-10 | 2012-04-03 | In Silico Biosciences | Method and apparatus for computer modeling of the interaction between and among cortical and subcortical areas in the human brain for the purpose of predicting the effect of drugs in psychiatric and cognitive diseases |
-
2023
- 2023-08-30 CN CN202311103673.5A patent/CN116825335B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113012815A (en) * | 2021-04-06 | 2021-06-22 | 西北工业大学 | Parkinson health risk assessment method based on multi-modal data |
| CN113963798A (en) * | 2021-10-15 | 2022-01-21 | 华中科技大学同济医学院附属协和医院 | Parkinson's disease motor symptom evaluation method |
| CN115762757A (en) * | 2022-11-09 | 2023-03-07 | 电子科技大学 | Prediction system for improving postoperative motion of Parkinson's disease based on individual characteristics |
| CN115762720A (en) * | 2022-11-17 | 2023-03-07 | 吾征智能技术(北京)有限公司 | System, method, electronic device and storage medium for assessing Parkinson's disease efficacy |
Non-Patent Citations (2)
| Title |
|---|
| Improved motor outcome prediction in Parkinson’s disease applying deep learning to DaTscan SPECT images;Matthew P. Adams;《Computers in Biology and Medicine 》;第1-8页 * |
| 基于人工智能算法的步态分析在疾病临床诊疗中的应用进展;刘芳超 等;《中国康复医学杂志》;第38卷(第06期);第836-839页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN116825335A (en) | 2023-09-29 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US20210057105A1 (en) | Method and apparatus for determining health status | |
| Wang et al. | MsTGANet: Automatic drusen segmentation from retinal OCT images | |
| CN116825335B (en) | Method and apparatus for predictive model training of prognostic outcome of parkinson's patient gait | |
| Narayanan et al. | Longitudinal falls-risk estimation using triaxial accelerometry | |
| CN108197592B (en) | Information acquisition method and device | |
| Muñoz-Organero et al. | Sensor optimization in smart insoles for post-stroke gait asymmetries using total variation and L 1 distances | |
| WO2021217906A1 (en) | Posture detection method, apparatus and device based on gait features, and storage medium | |
| CN116110597B (en) | Digital twinning-based intelligent analysis method and device for patient disease categories | |
| EP4411740A1 (en) | Model construction method and apparatus, device and medium | |
| CN115985515A (en) | Amblyopia correction effect prediction method, device and equipment based on machine learning | |
| CN114419676A (en) | Sitting posture analysis method and device based on artificial intelligence, computer equipment and medium | |
| Ekambaram et al. | Real-time AI-assisted visual exercise pose correctness during rehabilitation training for musculoskeletal disorder | |
| CN111481208B (en) | Auxiliary system, method and storage medium applied to joint rehabilitation | |
| CN116344042B (en) | Cognitive reserve intervention lifting method and system based on multi-modal analysis | |
| CN115268531B (en) | Water flow temperature regulation control method, device and equipment for intelligent bathtub and storage medium | |
| CN116206765A (en) | Device and storage medium for assessing risk level of parkinson's disease | |
| CN108550384B (en) | Method and device for pushing information | |
| CN115147768B (en) | Fall risk assessment method and system | |
| CN114373549B (en) | Self-adaptive exercise prescription health intervention method and system for old people | |
| Srikanth | Parkinson Disease Detection Using Various Machine Learning Algorithms | |
| CN116189898B (en) | Method and device for training predictive model for predicting parkinsonism risk level | |
| Li | An investigation of an athlete injury likelihood monitoring system using the random forest algorithm and DWT | |
| Ahmed et al. | ASAR Dataset and Computational Model for Affective State Recognition During ARAT Assessment for Upper Extremity Stroke Survivors | |
| CN117612724B (en) | Privacy assessment method and system for female vaginal relaxation state | |
| KR102421765B1 (en) | Control method, device and program of system for determining Pulmonary Thrombo-embolism based on non-contrast computed tomography images using artificial intelligence |
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 |