A machine learning-based retinal response prediction system

Abstract

The present invention discloses a retinal response prediction system based on machine learning, comprising: a data extraction module, configured to extract electroretinogram (ERG) data from a local data file, calculate relevant waveform parameters according to the ERG data, and organize the relevant waveform parameters to obtain an ERG analysis result classified by frequency; a feature engineering module, configured to merge the ERG analysis results at different frequencies, add light intensity information to the merged dataset, and clean the data; a model training module, configured to detect and remove outliers in the data by using the Isolation Forest algorithm, and respectively train a forward prediction model and a reverse prediction model according to the data by using different regression algorithms; a feature prediction module, configured to provide an interactive operation through a graphical user interface (GUI), obtain a data file according to a user instruction, and select a pre-trained machine learning model according to the instruction to perform forward or reverse prediction of retinal response, thereby improving the efficiency.

Description

Technical Field

[0001] The present invention relates to the field of retinal research, and particularly to a retinal response prediction system based on machine learning. Background Art

[0002] Since biological retinas are difficult to obtain, especially in humans and other higher animals, this limits the ability to conduct direct experimental research on retinal functions. Obtaining a sufficient number of retinal samples is a rather cumbersome task for related research.

[0003] The complexity and fragility of the retina require great caution in experimental operations. When studying the light stimulation variables of the retina, external conditions such as light conditions and equipment technology will interfere with the accuracy and reliability of the experiment. This increases the difficulty of related research to a certain extent.

[0004] In current visual prosthesis research, researchers believe that reconstructing light-stimulated images poses a great challenge. Some current deep learning models can reconstruct partial light-stimulated images based on the retinal spike signals of organisms, but these models run slowly, and it is difficult to obtain retinal spike signals, making it difficult to perform immediate light-stimulus reconstruction.

[0005] In current research on retinal signals, most focus on spike signals, and there is little research on retinal ERG. Researchers studying retinal ERG need to spend a lot of time manually sorting data, which greatly slows down the research progress. Summary of the Invention

[0006] To achieve the above and other related purposes, the present invention discloses a retinal response prediction system based on machine learning, including:

[0007] A data extraction module, configured to extract electroretinogram (ERG) data from a local data file, calculate relevant waveform parameters according to the ERG data, and organize the relevant waveform parameters to obtain an ERG analysis result classified by frequency;

[0008] A feature engineering module, configured to merge ERG analysis results at different frequencies, add light intensity information to the merged dataset, and clean the data;

[0009] A model training module, configured to detect and remove outliers in the data using the isolation forest algorithm, and train a forward prediction model and a reverse prediction model respectively according to the data by using different regression algorithms;

[0010] A feature prediction module, which is used to provide interactive operations through a graphical user interface (GUI), obtain a data file according to a user instruction, and select a pre-trained machine learning model according to the instruction to perform forward or reverse prediction of retinal responses.

[0011] Furthermore, the data extraction module includes:

[0012] An ERG information extraction module, which is used to extract electroretinogram (ERG) data from a local data file, calculate the latency, duration, amplitude, peak value of the a-wave, duration, amplitude, peak value of the b-wave, and save the above parameters to generate a parameter file;

[0013] A format conversion module, which is used to read the parameter file and convert the format of the parameter file;

[0014] A frequency segmentation module, which is used to segment the data in the parameter file after format conversion according to frequency.

[0015] Furthermore, the feature engineering module includes:

[0016] An added light intensity data module, which is used to supplement the light intensity information in the data;

[0017] A zero-value data deletion module, which is used to delete invalid data rows where both the a-wave amplitude and the b-wave amplitude in the data are zero;

[0018] An average value calculation module, which is used to read the cleaned data, group it according to frequency and light intensity, and calculate the average value of each group of data;

[0019] A file merging module, which is used to merge the average value with the data after deleting invalid data rows;

[0020] A file format conversion module, which is used to convert the merged file into a specified format;

[0021] A standard deviation calculation module, which is used to calculate the standard deviation of the data.

[0022] Furthermore, the model training module includes:

[0023] A forward integrated model training module, which is used to identify and filter outliers in the data using the Isolation Forest algorithm, perform regression prediction through stacking random forest and multi-layer perceptron neural network models, and use a custom accuracy calculation method to evaluate the prediction results, forming a model for predicting parameters in the electroretinogram according to the light frequency of the stimulating light and the light intensity of the stimulating light;

[0024] The forward MLP model training module is used to detect and remove outliers in the training data using the Isolation Forest algorithm, construct a Pipeline containing two StandardScalers and a Multi-Layer Perceptron neural network regressor, make predictions on the test set, calculate the prediction accuracy, and form a model for predicting the parameters in the electroretinogram based on the light frequency of the stimulating light and the light intensity of the stimulating light;

[0025] The forward random forest model training module is used to identify and filter outliers in the data using the Isolation Forest algorithm, and use GridSearchCV to optimize the hyperparameters of the polynomial features and the random forest regression model, make predictions on the test set, and calculate the accuracy, forming a model for predicting the parameters in the electroretinogram based on the light frequency of the stimulating light and the light intensity of the stimulating light;

[0026] The reverse MLP model training module is used to identify and filter outliers in the data using the Isolation Forest algorithm, and then establish a multi-layer perceptron regression model to form a model for predicting the light frequency and light intensity of the stimulating light based on the various parameters of the electroretinogram.

[0027] Furthermore, the model training module further includes:

[0028] The painting model training module is used to preprocess the features using Yeo-Johnson transformation and standardization, and use a multi-output gradient boosting regression model for training to obtain a painting model for generating waveform diagrams based on the predicted retinal response parameters.

[0029] Furthermore, the feature prediction module includes:

[0030] The graphical user interface module is used to receive user instructions in a human-computer interaction manner;

[0031] The retinal response reverse prediction module based on the MLP model is used to receive specific retinal response parameters, including latency, duration, amplitude, peak value of the a-wave, duration, amplitude, peak value of the b-wave, and predict the light intensity and frequency of the stimulating light through a pre-trained reverse MLP model;

[0032] The retinal response forward prediction program module based on the RF-MLP model is used to receive the frequency and light intensity of the stimulating light and predict the retinal response parameters through a pre-trained forward integrated model;

[0033] The retinal response forward prediction program module based on the random forest model is used to receive the frequency and light intensity of the stimulating light and predict the retinal response parameters through a pre-trained forward integrated model;

[0034] The forward prediction program module of the retinal response based on the MLP model is used to receive the frequency and intensity of the stimulating light, and make predictions through the pre-trained forward MLP model to obtain the retinal response parameters.

[0035] Further, the feature prediction module includes:

[0036] The retinal response waveform reconstruction module is used to receive the frequency and intensity of the stimulating light, select a specified training model according to the user's instruction to obtain the retinal response parameters, and call the drawing model to generate the retinal response waveform image.

[0037] Further, the feature prediction module includes:

[0038] The light stimulation reconstruction module is used to receive the retinal response parameters, obtain the frequency and intensity of the stimulating light according to the retinal response reverse prediction module based on the MLP model, and generate a light stimulation reduction video according to the frequency and intensity of the stimulating light.

[0039] By adopting the above technical solutions, first, the electroretinogram data is extracted from the local data file through the data extraction module, the relevant waveform parameters are calculated and sorted according to the electroretinogram data, then the light intensity information is added to the data set and the data is cleaned, then the model training based on machine learning is carried out according to the data, then the user's instruction is received according to the graphical user interface, and the forward or reverse prediction of the retinal response is carried out according to the user input data and the trained machine learning model, which speeds up the processing speed of the ERG data and improves the efficiency. Brief Description of the Drawings

[0040] In combination with the accompanying drawings and with reference to the following detailed description, the above and other features, advantages and aspects of the various embodiments of the present disclosure will become more apparent. The drawings are used to better understand the solution and do not limit the present disclosure. In the drawings, the same or similar reference numerals represent the same or similar elements, where:

[0041] Figure 1 It is the system block diagram of the embodiment of the present invention. Detailed Embodiments

[0042] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative efforts shall fall within the protection scope of the present invention.

[0043] Referring to Figure 1 , the embodiment of the present invention provides a retinal response prediction system based on machine learning, including:

[0044] A data extraction module, which is used to extract electroretinogram (ERG) data from a local data file, calculate relevant waveform parameters based on the ERG data, and organize the relevant waveform parameters to obtain an ERG analysis result classified by frequency.

[0045] Among them, the data extraction module includes:

[0046] An ERG information extraction module, which is used to extract electroretinogram (ERG) data from a local data file, calculate the latency, duration, amplitude, peak value of the a-wave, duration, amplitude, peak value of the b-wave, and save the above parameters to generate a parameter file.

[0047] Specifically, the preferred format of the local data file in this embodiment is an HDF5 format data file, and the parameter file is also an HDF5 format file.

[0048] A format conversion module, which is used to read the parameter file and convert the format of the parameter file.

[0049] Specifically, the format conversion module reads the data in the HDF5 file and converts it into a DataFrame object of Pandas. Then, the DataFrame object is exported and saved as an Excel file for further data processing and analysis.

[0050] A frequency segmentation module, which is used to segment the data in the parameter file after format conversion according to frequency.

[0051] Specifically, the frequency segmentation module segments the columns in the Excel file containing all frequency data according to a specific range, and saves the data in each frequency range to a separate Excel file, such as 1HZ.xlsx, 2HZ.xlsx, etc., for subsequent processing.

[0052] A feature engineering module, which is used to merge the ERG analysis results at different frequencies, add light intensity information to the merged dataset, and clean the data.

[0053] Among them, the feature engineering module includes:

[0054] A light intensity data adding module, which is used to supplement the light intensity information in the data.

[0055] As mentioned above, the light intensity information in the experimental data is supplemented, which is convenient for subsequent learning based on the data. The light intensity information is supplemented by those skilled in the art according to the actual data in the experimental process, and will not be elaborated here.

[0056] A zero-value data deletion module, which is used to delete the invalid data rows in the data where both the a-wave amplitude and the b-wave amplitude are zero.

[0057] As described above, the invalid data rows with both the a-wave amplitude and the b-wave amplitude being zero are deleted to eliminate the invalid data, reduce data redundancy, and improve data quality.

[0058] The average value calculation module is used to read the cleaned data, group the data according to frequency and light intensity, and calculate the average value of each group of data.

[0059] The file merging module is used to merge the average value with the data after deleting the invalid data rows.

[0060] The file format conversion module is used to convert the merged file into a specified format.

[0061] As described above, the merged Excel file is converted into a CSV format file to improve data accessibility and compatibility.

[0062] The standard deviation calculation module is used to calculate the standard deviation of the data.

[0063] The model training module is used to detect and remove outliers in the data using the Isolation Forest algorithm, and to train a forward prediction model and a reverse prediction model respectively according to the data by using different regression algorithms.

[0064] Among them, the model training module includes:

[0065] The forward integrated model training module is used to identify and filter outliers in the data using the Isolation Forest algorithm, perform regression prediction by stacking a random forest and a multi-layer perceptron neural network model, and use a custom accuracy calculation method to evaluate the prediction results, forming a model for predicting parameters in electroretinogram according to the light frequency of the stimulating light and the light intensity of the stimulating light.

[0066] As described above, the accuracy calculation method is set by those skilled in the art and will not be elaborated here.

[0067] The forward MLP model training module is used to detect and remove outliers in the training data using the Isolation Forest algorithm, construct a Pipeline containing two standard scalers and a multi-layer perceptron neural network regressor, make predictions on the test set, calculate the prediction accuracy, and form a model for predicting parameters in electroretinogram according to the light frequency of the stimulating light and the light intensity of the stimulating light.

[0068] The forward random forest model training module is used to identify and filter outliers in the data using the Isolation Forest algorithm, use GridSearchCV to optimize the hyperparameters of the polynomial features and the random forest regression model, make predictions on the test set, and calculate the accuracy, forming a model for predicting parameters in electroretinogram according to the light frequency of the stimulating light and the light intensity of the stimulating light.

[0069] The reverse MLP model training module is used to identify and filter outliers in the data using the Isolation Forest algorithm, and then establish a multi-layer perceptron regression model to form a model for predicting the light frequency and light intensity of the stimulating light based on the various parameters of the electroretinogram.

[0070] The painting model training module is used to preprocess the features using Yeo-Johnson transformation and normalization, and train using a multi-output gradient boosting regression model to obtain a painting model for generating a waveform diagram based on the predicted retinal response parameters.

[0071] As mentioned above, the specific model training method is a method well-known to those skilled in the art and will not be elaborated here.

[0072] The feature prediction module is used to provide an interactive operation through a graphical user interface (GUI), obtain a data file according to a user instruction, and perform forward or reverse prediction of the retinal response according to the instruction by selecting a pre-trained machine learning model.

[0073] Among them, the feature prediction module includes:

[0074] The graphical user interface module is used to receive user instructions in a human-computer interaction manner.

[0075] Specifically, the system provides an interactive operation through a graphical user interface (GUI), allowing the user to upload a data file in Excel or HDF5 format, select a pre-trained machine learning model, and perform forward and reverse prediction of the retinal response.

[0076] The retinal response reverse prediction module based on the MLP model is used to receive specific retinal response parameters, including latency, duration, amplitude, peak value of the a-wave, duration, amplitude, peak value of the b-wave, and predict through a pre-trained reverse MLP model to obtain the light intensity and frequency of the stimulating light.

[0077] Among them, when this prediction module obtains the retinal response parameters, it can be manually input through an input box or obtained by reading an Excel data file containing the retinal response parameters.

[0078] The retinal response forward prediction program module based on the RF-MLP model is used to receive the frequency and light intensity of the stimulating light and predict through a pre-trained forward integration model to obtain the retinal response parameters.

[0079] The retinal response forward prediction program module based on the random forest model is used to receive the frequency and light intensity of the stimulating light and predict through a pre-trained forward integration model to obtain the retinal response parameters.

[0080] The forward prediction program module of retinal response based on the MLP model is used to receive the frequency and intensity of the stimulating light, and make predictions through a pre-trained forward MLP model to obtain retinal response parameters.

[0081] As mentioned above, each module can select local and global stimulation modes according to the actual needs of users, and the mode settings are set by those skilled in the art according to actual needs.

[0082] The retinal response waveform reconstruction module is used to receive the frequency and intensity of the stimulating light, select a specified training model according to the user's instruction to obtain retinal response parameters, and call a painting model to generate a retinal response waveform image.

[0083] As mentioned above, the frequency and intensity of the stimulating light can be manually input by those skilled in the art, or obtained by reading an Excel file pre-stored with light frequency and intensity data.

[0084] The light stimulation reconstruction module is used to receive retinal response parameters, obtain the frequency and intensity of the stimulating light according to the retinal response inverse prediction module based on the MLP model, and generate a light stimulation reduction video according to the frequency and intensity of the stimulating light.

[0085] As mentioned above, retinal response parameters can be manually input by those skilled in the art, or obtained by reading an Excel file pre-stored with retinal response parameter data.

[0086] The embodiment of the present invention provides a retinal response prediction system based on machine learning. First, the electroretinogram data is extracted from a local data file through a data extraction module, relevant waveform parameters are calculated and sorted according to the electroretinogram data, then light intensity information is added to the dataset and the data is cleaned, then model training based on machine learning is performed according to the data, then user instructions are received according to the graphical user interface, and forward or inverse prediction of retinal response is performed according to the data input by the user and the trained machine learning model, which speeds up the processing speed of ERG data and improves the efficiency.

[0087] Those skilled in the art of this technology can understand that, unless otherwise defined, all terms used here (including technical terms and scientific terms) have the same meaning as the general understanding of those of ordinary skill in the art to which the present invention belongs. It should also be understood that terms such as those defined in a general dictionary should be understood to have a meaning consistent with the meaning in the context of the prior art, and will not be interpreted with an idealized or overly formal meaning unless specifically defined.

[0088] For method embodiments, for the sake of simple description, they are all expressed as a series of action combinations. However, those skilled in the art should be aware that the embodiments of the present invention are not limited by the described action sequences, because according to the embodiments of the present invention, certain steps can be performed in other sequences or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions involved are not necessarily essential to the embodiments of the present invention.

[0089] From the description of the above embodiments, those skilled in the art can clearly understand that this application can be implemented by means of software plus a necessary general hardware platform. Based on such an understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in various embodiments or some parts of the embodiments of this application.

[0090] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A retinal response prediction system based on machine learning, characterized in that: include: The data extraction module is used to extract electroretinogram (ERG) data from the local data file, calculate relevant waveform parameters based on the ERG data, and sort the relevant waveform parameters to obtain ERG analysis results classified by frequency; Feature engineering module, used to merge ERG analysis results at different frequencies, add light intensity information to the merged data set, and clean the data; The model training module is used to detect and remove outliers in the data using the isolation forest algorithm, and to train the forward prediction model and the reverse prediction model based on the data by using different regression algorithms; The model training module includes: A forward ensemble model training module is used to identify and filter outliers in the data using an isolation forest algorithm, perform regression predictions by stacking random forest and multilayer perceptron neural network models, and use a custom accuracy calculation method to evaluate the prediction results, forming a model that predicts parameters in the electroretinogram based on the light frequency and light intensity of the stimulus light; The forward MLP model training module is used to detect and remove outliers in the training data using the isolation forest algorithm, build a pipeline containing two standard scalers and a multi-layer perceptron neural network regressor, and make predictions on the test set, calculate the prediction accuracy, and form a model that predicts the parameters in the electroretinogram according to the light frequency and light intensity of the stimulus light; The forward random forest model training module is used to use the isolation forest algorithm to identify and filter outliers in the data, and use grid search CV to optimize the hyperparameters of the polynomial features and the random forest regression model, make predictions on the test set, and calculate the accuracy, forming a model that predicts the parameters in the electroretinogram based on the light frequency and light intensity of the stimulus light; The reverse MLP model training module is used to identify and filter outliers in the data using the isolation forest algorithm, and then establish a multi-layer perception regression model to form a model that predicts the light frequency and light intensity of the stimulus light based on the various parameters of the electroretinogram The feature prediction module is used to provide interactive operations through a graphical user interface (GUI), obtain data files according to user instructions, and select a pre-trained machine learning model according to the instructions to perform forward or reverse prediction of retinal response.

2. The system according to claim 1, characterized in that The data extraction module comprises: ERG information extraction module, used to extract electroretinogram (ERG) data from local data files, calculate latency, duration, amplitude, peak value of a wave, duration, amplitude, peak value of b wave, save the above parameters and generate parameter files; A format conversion module is used to read parameter files and convert the format of the parameter files; The frequency segmentation module is used to segment the data in the parameter file after format conversion according to frequency.

3. The system according to claim 2, characterized in that The feature engineering module includes: Added light intensity data module to supplement the light intensity information in the data; The zero value data deletion module is used to delete invalid data rows in which both the a-wave amplitude and the b-wave amplitude are zero; The average calculation module is used to read the cleaned data, group them according to frequency and light intensity, and calculate the average value of each group of data; A file merging module, used to merge the average value with the data after deleting invalid data rows; A file format conversion module, used to convert the merged file into a specified format; The standard deviation calculation module is used to calculate the standard deviation of data.

4. The system according to claim 1, characterized in that The model training module also includes: The painting model training module is used to preprocess the features using Yeo-Johnson transformation and normalization, and train using a multi-output gradient boosting regression model to obtain a painting model that generates a waveform diagram based on the predicted retinal response parameters.

5. The system according to claim 1, characterized in that The feature prediction module comprises: A graphical user interface module, used for receiving user instructions in a human-computer interaction manner; The retinal response reverse prediction module based on the MLP model is used to receive specific retinal response parameters, including latency, duration, amplitude, peak value of a wave, duration, amplitude, peak value of b wave, and predict through the pre-trained reverse MLP model to obtain the intensity and frequency of the stimulus light; The retinal response forward prediction program module based on the RF-MLP model is used to receive the frequency and intensity of the stimulus light, and predict the retinal response parameters through the pre-trained forward integration model; The retinal response forward prediction program module based on the random forest model is used to receive the frequency and intensity of the stimulus light, and predict the retinal response parameters through the pre-trained forward integration model; The retinal response forward prediction program module based on the MLP model is used to receive the frequency and intensity of the stimulus light, and predict it through the pre-trained forward MLP model to obtain the retinal response parameters.

6. The system according to claim 4, characterized in that The feature prediction module comprises: The retinal response waveform reconstruction module is used to receive the frequency and intensity of the stimulus light, select the specified training model according to the user's instructions to obtain the retinal response parameters, and call the painting model to generate the retinal response waveform image.

7. The system according to claim 5, characterized in that The feature prediction module comprises: The light stimulation reconstruction module is used to receive retinal response parameters, obtain the frequency and intensity of the stimulation light according to the retinal response reverse prediction module based on the MLP model, and generate a light stimulation restoration video according to the frequency and intensity of the stimulation light.

Images