Intelligent recognition system for cardiac amyloidosis based on ultrasonic and electrocardiogram data

Abstract

The application provides an intelligent recognition system and method for cardiac amyloidosis based on ultrasonic and electrocardiogram data, and the system comprises an electrocardio-ultrasound recognition module, a voltage index calculation module, a parameter intelligent classification module and a SHAP analysis module. Based on the deep learning architecture, the technology is based on multi-modal data, based on electrocardiogram and echocardiogram, and multi-modal information fusion is carried out on the rare disease patient case image information encountered in clinical diagnosis. Parallel operation is adopted for key feature extraction and recognition, which can efficiently and reliably realize the classification of cardiac amyloidosis disease and provide strong support for rapid diagnosis of doctors.

Description

Technical Field

[0001] This invention belongs to the field of intelligent medical care, specifically relating to an intelligent identification system for cardiac amyloidosis based on ultrasound and electrocardiogram data. Background Technology

[0002] Cardiac amyloidosis is an idiopathic cardiomyopathy, mainly manifested as heart failure, arrhythmias, and other related symptoms. Current mainstream diagnostic techniques rely on examinations such as electrocardiograms and echocardiograms performed by clinicians.

[0003] With breakthroughs in deep learning algorithms, big data processing capabilities, and cloud computing technologies, intelligent technologies have made significant progress in fields such as image recognition and natural language processing. Combining intelligent technologies with medical diagnosis has enabled applications such as automatic localization of lung nodules and tumors in imaging and identification of cancer cells in pathological slide analysis. Addressing the complexities of analyzing various types of pathological data, particularly for cardiac amyloidosis, a rare disease, this paper proposes an intelligent identification system and method for cardiac amyloidosis based on electrocardiogram (ECG) and echocardiography. This system utilizes a deep learning architecture, leverages multimodal data, and aims to improve the diagnostic efficiency of frontline physicians. Summary of the Invention

[0004] The purpose of this invention is to provide an intelligent identification system and method for cardiac amyloidosis based on ultrasound and electrocardiogram (ECG) data. Based on ECG and echocardiography, multimodal information fusion is performed on image information of rare disease patients encountered in clinical diagnosis. Parallel computing is used for key feature extraction and identification, which can efficiently and reliably classify cardiac amyloidosis, providing strong support for doctors' rapid diagnosis.

[0005] This invention provides an intelligent identification system for cardiac amyloidosis based on ultrasound and electrocardiogram data. The system includes: an electrocardiogram and echocardiogram identification module, a voltage index calculation module, a parameter intelligent classification module, and a SHAP analysis module.

[0006] Furthermore, the ECG and echocardiogram recognition module consists of a waveform contour detection module and an OCR recognition module; the voltage index calculation module consists of a parameter extraction module and a voltage index calculation module; the parameter intelligent classification module consists of a model training module and a disease classification module; and the SHAP analysis module consists of a SHAP plotting module and a model interpretability analysis module.

[0007] The ECG and echocardiogram recognition module receives the 12-lead ECG and echocardiogram report from the patient. The waveform contour detection module uses the 12-lead ECG as input, performs image analysis on key waveforms, extracts key voltages, and notifies the voltage index calculation module. The OCR recognition module uses the echocardiogram report as input, performs OCR text recognition on key parameters, and simultaneously transmits the recognition results to the voltage index calculation module.

[0008] After receiving the electrocardiogram (ECG) information, the waveform contour detection module first normalizes each waveform to remove redundant information and enhance key information, obtaining a preprocessed result for each waveform. Next, convolution filtering is performed to extract image features and obtain image gradient information, thereby identifying the magnitude of extreme values ​​in each waveform.

[0009] The workflow of the waveform contour detection module is as follows:

[0010] 1. Perform grayscale conversion, smoothing filtering, and contrast enhancement on the input 12-lead electrocardiogram;

[0011] 2. Edge detection is performed on the waveforms of each lead by fusing brightness and texture features;

[0012] 3. Refine the edges along the gradient direction, retaining only the local maximum response points, and complete the edge post-processing operation;

[0013] 4. Extract closed waveform contours from the binary edge map;

[0014] 5. Quantify waveform contour features, perform extreme value analysis on key waveforms, and calculate key features;

[0015] 6. Notify the voltage calculation module of the key features calculated in step 5.

[0016] The workflow of the voltage index calculation module is as follows:

[0017] 1. Receive key parameters output by the ECG and echocardiography recognition module;

[0018] 2. Complete the calculation of the sokolow-lyon voltage to ECG voltage quality ratio;

[0019] 3. Based on the calculation results of step 2, determine whether the results conform to the characteristics of cardiac amyloidosis;

[0020] 4. Based on the judgment results in step 3, if the results match, the key characteristics of the case are notified to the intelligent classification module; otherwise, the auxiliary diagnosis results are output directly.

[0021] The workflow of the parameter intelligent classification module is as follows:

[0022] 1. Using the key parameters output from the previous module as samples, load the data and divide it into training and test sets;

[0023] 2. Iterative training is performed through operations such as calculating gradients and second derivatives, constructing decision trees, calculating leaf weights, and updating the model;

[0024] 3. Set a termination condition and make a judgment; stop iterating when the preset number of trees is reached.

[0025] 4. Calculate metrics such as accuracy using the test set, and adjust the regularization parameters to optimize the model;

[0026] 5. Traverse all tree structures for the test samples, accumulate the weights of the leaf nodes to obtain the final prediction, complete the classification of cardiac amyloidosis types, and support the identification of different disease types such as AL, HCM, and ATTR.

[0027] The workflow of the SHAP analysis module is as follows:

[0028] 1. Define the global behavior analysis mode of the model based on the key parameters output by the intelligent classification module;

[0029] 2. Perform data preprocessing, handle missing values, standardize feature encoding, and retain an independent validation set for interpreting stability tests;

[0030] 3. Choose Tree SHAP as the interpretation method, and ensure that the model's predictive performance meets the standards before interpretation;

[0031] 4. Complete model iteration;

[0032] 5. Draw waterfall plots and force maps to show the feature contribution paths and feature pair prediction directions for individual predictions;

[0033] 6. Output visualization results.

[0034] The beneficial effects of this invention are as follows:

[0035] 1. The intelligent identification technology for cardiac amyloidosis provided by this invention can utilize existing case databases and leverage the advantages of intelligent algorithms to meet the requirements of high speed, real-time performance, and accuracy in assisted diagnosis;

[0036] 2. This invention designs an electrocardiogram (ECG) recognition technology based on image contour detection, which analyzes twelve-lead ECG data. It can make full use of the waveform texture information of each lead ECG, remove redundant image features, make image information more concentrated, and improve the stability and robustness of feature extraction.

[0037] 3. This invention designs a cardiac ultrasound report extraction technology based on OCR recognition, which uses OCR recognition technology to extract key parameters from cardiac ultrasound reports;

[0038] 4. This invention designs an intelligent classification technology based on XGBoost boosting tree, which makes full use of key parameters for intelligent classification and completes the inference of the probability of cardiac amyloidosis type. At the same time, SHAP is used to analyze the influence of each feature on the model, which effectively improves the interpretability of the model.

[0039] 5. This invention significantly improves the diagnostic efficiency of doctors in treating cardiac amyloidosis, a rare cardiomyopathy. The XGBoost model trained by this invention achieves a 97% accuracy rate in classifying case data, enhancing the ability to rapidly analyze and integrate complex information. Through advanced algorithms and data analysis techniques, doctors can more accurately identify early signs and characteristics of the disease, thereby enabling earlier diagnosis and intervention. This invention aims to fully utilize the advantages of modern technology through a "medical + intelligent" approach to further advance the field of rare disease research and improve the quality and efficiency of medical services. Attached Figure Description

[0040] The accompanying drawings, which form part of this application, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0041] Figure 1 This is a system structure diagram of the present invention;

[0042] Figure 2 This is a flowchart of the waveform contour detection module.

[0043] Figure 3 This is a flowchart of the voltage index calculation module.

[0044] Figure 4 This is a flowchart of the parameter intelligent classification module.

[0045] Figure 5 This is a flowchart of the SHAP analysis module workflow. Detailed Implementation

[0046] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0047] This invention provides an intelligent identification system for cardiac amyloidosis based on ultrasound and electrocardiogram data. The system includes: an electrocardiogram and echocardiogram identification module, a voltage index calculation module, a parameter intelligent classification module, and a SHAP analysis module.

[0048] Furthermore, the ECG and echocardiogram recognition module consists of a waveform contour detection module and an OCR recognition module; the voltage index calculation module consists of a parameter extraction module and a voltage index calculation module; the parameter intelligent classification module consists of a model training module and a disease classification module; and the SHAP analysis module consists of a SHAP plotting module and a model interpretability analysis module.

[0049] The ECG and echocardiogram recognition module receives the 12-lead ECG and echocardiogram report from the patient. The waveform contour detection module uses the 12-lead ECG as input, performs image analysis on key waveforms, extracts key voltages, and notifies the voltage index calculation module. The OCR recognition module uses the echocardiogram report as input, performs OCR text recognition on key parameters, and simultaneously transmits the recognition results to the voltage index calculation module.

[0050] After receiving the electrocardiogram (ECG) information, the waveform contour detection module first normalizes each waveform to remove redundant information and enhance key information, obtaining a preprocessed result for each waveform. Next, convolution filtering is performed to extract image features and obtain image gradient information, thereby identifying the magnitude of extreme values ​​in each waveform.

[0051] The workflow of the waveform contour detection module is as follows:

[0052] 1. Perform grayscale conversion, smoothing filtering, and contrast enhancement on the input 12-lead electrocardiogram;

[0053] 2. Edge detection is performed on the waveforms of each lead by fusing brightness and texture features;

[0054] 3. Refine the edges along the gradient direction, retaining only the local maximum response points, and complete the edge post-processing operation;

[0055] 4. Extract closed waveform contours from the binary edge map;

[0056] 5. Quantify waveform contour features, perform extreme value analysis on key waveforms, and calculate key features;

[0057] 6. Notify the voltage calculation module of the key features calculated in step 5.

[0058] The workflow of the voltage index calculation module is as follows:

[0059] 1. Receive key parameters output by the ECG and echocardiography recognition module;

[0060] 2. Complete the calculation of the sokolow-lyon voltage to ECG voltage quality ratio;

[0061] 3. Based on the calculation results of step 2, determine whether the results conform to the characteristics of cardiac amyloidosis;

[0062] 4. Based on the judgment results in step 3, if the results match, the key characteristics of the case are notified to the intelligent classification module; otherwise, the auxiliary diagnosis results are output directly.

[0063] The workflow of the parameter intelligent classification module is as follows:

[0064] 1. Using the key parameters output from the previous module as samples, load the data and divide it into training and test sets;

[0065] 2. Iterative training is performed through operations such as calculating gradients and second derivatives, constructing decision trees, calculating leaf weights, and updating the model;

[0066] 3. Set a termination condition and make a judgment; stop iterating when the preset number of trees is reached.

[0067] 4. Calculate metrics such as accuracy using the test set, and adjust the regularization parameters to optimize the model;

[0068] 5. Traverse all tree structures for the test samples, accumulate the weights of the leaf nodes to obtain the final prediction, complete the classification of cardiac amyloidosis types, and support the identification of different disease types such as AL, HCM, and ATTR.

[0069] The workflow of the SHAP analysis module is as follows:

[0070] 1. Define the global behavior analysis mode of the model based on the key parameters output by the intelligent classification module;

[0071] 2. Perform data preprocessing, handle missing values, standardize feature encoding, and retain an independent validation set for interpreting stability tests;

[0072] 3. Choose Tree SHAP as the interpretation method, and ensure that the model's predictive performance meets the standards before interpretation;

[0073] 4. Complete model iteration;

[0074] 5. Draw waterfall plots and force maps to show the feature contribution paths and feature pair prediction directions for individual predictions;

[0075] 6. Output visualization results.

[0076] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A smart identification system for cardiac amyloidosis based on ultrasound and electrocardiogram data, characterized in that, include: ECG and echocardiography recognition module, voltage index calculation module, intelligent parameter classification module, and SHAP analysis module; The ECG and echocardiogram recognition module consists of a waveform contour detection module and an OCR recognition module; it is used to receive 12-lead ECG and echocardiogram reports. The voltage index calculation module consists of a parameter extraction module and a voltage index calculation module; The parameter intelligent classification module consists of a model training module and a disease classification module; The SHAP analysis module consists of a SHAP plotting module and a model interpretability analysis module.

2. The intelligent identification system for cardiac amyloidosis based on ultrasound and electrocardiogram data according to claim 1, characterized in that, The waveform contour detection module uses a twelve-lead electrocardiogram as input to perform image analysis on key waveforms, extract key voltages, and notify the voltage index calculation module.

3. The intelligent identification system for cardiac amyloidosis based on ultrasound and electrocardiogram data according to claim 2, characterized in that, The OCR recognition module takes the cardiac ultrasound report as input, performs OCR text recognition on key parameters, and simultaneously transmits the recognition results to the voltage index calculation module.

4. The intelligent identification system for cardiac amyloidosis based on ultrasound and electrocardiogram data according to claim 2, characterized in that, After receiving the electrocardiogram information, the waveform contour detection module first normalizes each waveform to remove redundant information and enhance key information, thus obtaining the preprocessing result of each waveform. Next, it performs convolution filtering to extract image features, obtain image gradient information, and then identifies the magnitude of the extreme values ​​in each waveform.

5. The intelligent identification system for cardiac amyloidosis based on ultrasound and electrocardiogram data according to claim 4, characterized in that, The working steps of the waveform contour detection module are as follows: Step 1: Perform grayscale conversion, smoothing filtering, and contrast enhancement on the input 12-lead electrocardiogram; Step 2: Perform edge detection on the waveforms of each lead by fusing brightness and texture features; Step 3: Refine the edges along the gradient direction, retaining only the local maximum response points, to complete the edge post-processing operation; Step 4: Extract the closed waveform profile from the binary edge map; Step 5: Quantify waveform contour features, perform extreme value analysis on key waveforms, and calculate key features; Step 6: Notify the voltage calculation module of the key features calculated in Step 5.

6. The intelligent identification system for cardiac amyloidosis based on ultrasound and electrocardiogram data according to claim 4, characterized in that, The workflow of the voltage index calculation module is as follows: Step 1: Receive key parameters output by the ECG and echocardiography recognition module; Step 2: Complete the calculation of the sokolow-lyon voltage to ECG voltage quality ratio; Step 3: Based on the calculation results of Step 2, determine whether the results meet the characteristics of cardiac amyloidosis. Step 4: Based on the judgment results in Step 3, if the results match, notify the intelligent classification module of the key characteristics of the case; otherwise, directly output the auxiliary diagnosis results.

7. The intelligent identification system for cardiac amyloidosis based on ultrasound and electrocardiogram data according to claim 4, characterized in that, The workflow of the parameter intelligent classification module is as follows: Step 1: Use the key parameters output by the previous module as samples, load the data, and divide it into training and test sets; Step 2: Iterative training is performed by calculating gradients and second derivatives, constructing decision trees, calculating leaf weights, and updating the model. Step 3: Set the termination condition and make a judgment. Stop the iteration when the preset number of trees is reached. Step 4: Calculate metrics such as accuracy using the test set, and adjust the regularization parameters to optimize the model; Step 5: Traverse all tree structures for the test samples, accumulate the weights of the leaf nodes to obtain the final prediction, complete the classification of cardiac amyloidosis types, and support the identification of different disease types such as AL, HCM, and ATTR.

8. The intelligent identification system for cardiac amyloidosis based on ultrasound and electrocardiogram data according to claim 4, characterized in that, The workflow of the SHAP analysis module is as follows: Step 1: Define the global behavior analysis mode of the model based on the key parameters output by the intelligent classification module; Step 2: Perform data preprocessing, handle missing values, standardize feature encoding, and retain an independent validation set for interpreting stability tests; Step 3: Select Tree SHAP as the interpretation method, and ensure that the model's predictive performance meets the requirements before interpretation; Step 4: Complete model iteration; Step 5: Draw a waterfall plot and force map to show the feature contribution path and feature pair prediction direction of a single prediction; Step 6: Output the visualization results.

Images