Coronary heart disease early warning model construction method and system based on auricular point multi-modal characteristics

Abstract

The invention relates to the field of medical information processing, in particular to a coronary heart disease early warning model construction method and system based on auricular point multi-modal features. The method comprises the following steps: using ear region temperature features, ear region blood flow features, ear region morphological features and traditional risk factors as input of a machine learning model, using whether the patient suffers from the coronary heart disease as a label, training the machine learning model, and obtaining an early warning model of the coronary heart disease. The traditional Chinese medicine auricular point theory and the modern machine learning technology are combined for the first time, correlation between auricular point characteristics and coronary heart disease incidence is verified through prospective research, a scientific basis is provided for prevention of diseases, and application of combination of traditional Chinese medicine and western medicine in the field of cardiovascular disease prevention is promoted.

Description

Technical Field

[0001] This invention relates to the field of medical information processing, specifically to a method and system for constructing an early warning model for coronary heart disease based on multimodal features of auricular acupoints. Background Technology

[0002] Coronary atherosclerotic heart disease (CAD) is one of the leading causes of death and disability among cardiovascular diseases worldwide. According to the "China Cardiovascular Health and Disease Report 2023," the number of people with CAD in my country exceeds 11 million, and the incidence rate is trending towards younger ages. Early pathological changes in CAD (such as coronary artery microstenosis and endothelial dysfunction) lack typical clinical symptoms. By the time symptoms such as chest pain and chest tightness appear, the disease has often progressed to the middle or late stages, missing the optimal intervention window.

[0003] Currently used clinical methods for early screening of coronary heart disease have significant limitations:

[0004] 1) Invasive / High-cost technologies: Although coronary angiography (the gold standard for diagnosis) and coronary CT angiography (CCT) are highly accurate, they have risks of invasive operation (such as puncture complications and contrast agent nephropathy), radiation exposure, and high cost (the cost of a single examination exceeds 2,000 yuan), which cannot meet the needs of large-scale screening of asymptomatic populations;

[0005] 2) Limitations of traditional non-invasive techniques: Traditional non-invasive methods such as electrocardiogram and exercise stress test have low sensitivity (about 50%-60%), making it difficult to detect early subtle lesions; early warning models based on facial features (CN111816308B) and fundus images (CN202311567418.6) rely only on indirect features of the body surface and do not associate with auricular acupoints that are directly related to cardiovascular function, resulting in insufficient specificity of early warning.

[0006] 3) Bottlenecks in TCM Auricular Diagnosis: In TCM theory, "the ear is where all the meridians converge," and the physiological changes (temperature, blood flow, morphology) of the "heart area" and surrounding regions of the ear are closely related to heart function, a typical manifestation of "what is inside will inevitably manifest outside." However, traditional auricular diagnosis relies on the physician's subjective experience (such as visual observation of morphology and finger palpation of temperature), lacks objective quantitative standards, and has not formed a clear "feature-risk" correlation model, making it difficult to promote and apply.

[0007] 4) Deficiencies of existing models: The publicly available coronary heart disease prediction models (such as the improved random forest model in CN202110488133.8 and the logistic regression model in CN202211386274.X) mostly use clinical biochemical indicators (such as LDL-C and hs-CRP) as inputs, do not integrate the multimodal features of auricular acupoints, and cannot use the advantages of traditional Chinese medicine theory to achieve "early-non-invasive-precise" early warning.

[0008] In summary, existing technologies have not yet solved the core problems of "non-invasive early screening of coronary heart disease, objectification of auricular acupoint features, and integration of traditional Chinese and Western medicine in early warning models." There is an urgent need to build a set of early warning technology solutions for coronary heart disease based on multimodal quantitative features of auricular acupoints and combined with machine learning. Summary of the Invention

[0009] In view of the shortcomings of the prior art, the core objective of this invention is to solve the following technical problems:

[0010] 1) Breaking through the subjective limitations of traditional auricular acupoint diagnosis, we use standardized technologies (infrared thermal imaging, laser speckle, etc.) to quantify the temperature, blood flow, and morphological characteristics of auricular acupoints, and establish an objective auricular acupoint characteristic evaluation system.

[0011] 2) To overcome the deficiency of existing early warning models in not integrating multimodal features of auricular acupoints, a fusion input model of "multimodal features of auricular acupoints + traditional risk factors" is constructed to improve the sensitivity and specificity of early warning of coronary heart disease;

[0012] 3) To address the issues of convenience and cost-effectiveness in large-scale population screening, a non-invasive, low-cost, and easy-to-operate early warning solution is provided, suitable for primary healthcare institutions and community settings;

[0013] 4) Provide modern technical support for the TCM theory of "prevention of disease", verify the temporal correlation between auricular acupoint characteristics and the onset of coronary heart disease through prospective research, and clarify the evolution law of auricular acupoint characteristics from "pre-disease - impending disease - disease".

[0014] Ultimately, the goal is to provide an early warning model and system for coronary heart disease based on multimodal features of auricular acupoints. This system can quickly assess the early risk of coronary heart disease without invasive procedures or specialized and complex equipment, providing a basis for early clinical intervention.

[0015] This invention provides the following technical solution:

[0016] A method for constructing an early warning model for coronary heart disease based on multimodal features of auricular acupoints includes the following steps:

[0017] Using ear region temperature characteristics, ear region blood flow characteristics, ear region morphological characteristics, and traditional risk factors as inputs to a machine learning model, and using whether or not someone has coronary heart disease as a label, the machine learning model is trained to obtain an early warning model for coronary heart disease.

[0018] Preferably, the traditional risk factors are age, sex, low-density lipoprotein cholesterol, high-sensitivity C-reactive protein, and smoking history.

[0019] Preferably, the ear region temperature characteristic is as follows: The blood flow characteristics of the ear region are as follows: The morphological features of the ear region ;

[0020] in, This indicates the temperature difference between the cardiac region and the concha cymbidium region; This indicates the average blood perfusion in the cardiac region;

[0021] This represents the rate of change in blood perfusion relative to healthy individuals; the calculation method is as follows:

[0022] in, This represents the reference value for average blood perfusion in healthy individuals.

[0023] The score represents the evaluation of the ear region morphology from four dimensions: papules / nodules, vascular abnormalities, pigmentation, and skin depressions / protrusions.

[0024] Preferably, the label of whether or not one has coronary artery disease is determined by the Gensini score of the degree of stenosis on coronary angiography.

[0025] Preferably, the machine learning algorithm includes one or more ensemble models selected from Random Forest, Support Vector Machine, and ResNet-50.

[0026] Preferably, the hyperparameters of the random forest are: number of decision trees = 500, maximum number of features = 5, minimum number of sample splits = 10; the hyperparameters of the support vector machine are: kernel function is radial basis kernel, penalty coefficient C = 10, gamma = 0.1; the hyperparameters of the ResNet-50 are: learning rate = 0.001, number of iterations = 100, dropout ratio = 0.3.

[0027] This invention also provides a method for early warning of coronary heart disease based on multimodal features of auricular acupoints. The method is characterized by using ear temperature features, ear blood flow features, ear morphological features, and traditional risk factors as inputs to an early warning model for coronary heart disease to obtain a coronary heart disease warning risk score. The early warning model for coronary heart disease is the aforementioned early warning model for coronary heart disease.

[0028] The present invention also provides an electronic device system,

[0029] include:

[0030] Memory, used to store computer programs;

[0031] A processor, when executing a program stored in memory, implements the steps of the method according to any one of claims 1 to 7.

[0032] The present invention also provides a computer program product, including a computer program, characterized in that the computer program implements the above-described method when executed by a processor.

[0033] Beneficial effects:

[0034] 1) Non-invasive and accurate early warning: The ear acupoint features are collected using non-invasive technologies such as infrared thermal imaging and laser speckle, avoiding the risks of invasive examinations; the integrated model has an AUC of 0.89 and a sensitivity of 0.82 for early identification of high-risk groups, which is significantly better than existing technologies;

[0035] 2) Objective Quantitative Breakthrough: Establish quantitative standards for multimodal features of auricular acupoints and a positive judgment system to solve the subjective problem of traditional auricular acupoint diagnosis and provide objective technical support for TCM auricular acupoint diagnosis;

[0036] 3) Economical, convenient and applicable: The system equipment is portable, the cost of a single screening is less than 200 yuan, and no professional physicians are required to operate it. It is suitable for large-scale population screening in communities and primary healthcare institutions.

[0037] 4) Integration and Innovation of Traditional Chinese and Western Medicine: For the first time, the theory of auricular acupuncture in traditional Chinese medicine is combined with modern machine learning technology. Through prospective research, the correlation between auricular acupuncture characteristics and the onset of coronary heart disease is verified, providing a scientific basis for "prevention of disease" and promoting the application of integrated traditional Chinese and Western medicine in the field of cardiovascular disease prevention. Detailed Implementation

[0039] To make the objectives, features, and advantages of this invention more apparent and understandable, the technical solutions in the embodiments of this invention will be clearly and completely described below. Obviously, the embodiments described below are only a part of the embodiments of this invention, and not all of them. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0040] (I) Acquisition and Quantification of Multimodal Features of Ear Acupoints

[0041] Based on the theory of the connection between auricular acupoints and the cardiovascular system in Traditional Chinese Medicine, this study focuses on the "heart area" auricular acupoint (located in the central depression of the concha according to the standard GB / T 13734-2009 "Names and Locations of Auricular Acupoints") and its surrounding related areas (earlobe fold area). Three standardized techniques were used to collect multimodal features:

[0042] 1. Infrared thermal imaging temperature feature acquisition

[0043] Equipment: High-resolution infrared thermal imager (such as Hikvision H21Pros, temperature measurement range -20℃~550℃, spatial resolution 256×192);

[0044] Data collection process: Subjects sit quietly for 15 minutes (to avoid interference from ambient temperature), remove ear ornaments, use a white, non-reflective background as the collection environment, point the lens vertically at the ear at a fixed distance of 30cm, and collect infrared thermal images of the "heart area" of both ears and the earlobe fold area.

[0045] Quantitative indicators: , , ;

[0046] in, This indicates the average temperature of the heart region;

[0047] This indicates the temperature difference between the cardiac region and the concha cymbidium region;

[0048] Indicates the heart region.

[0049] 2. Laser speckle blood flow characteristic acquisition

[0050] Equipment: Laser speckle (RFLSI PRO, Shenzhen Ruiwode Life Science Co., Ltd.);

[0051] Data collection procedure: The subject keeps their ear relaxed, the probe is aligned with the skin of the "heart area", and data is collected continuously for 30 seconds at a sampling frequency of 10Hz;

[0052] Quantitative indicators: , , , ;

[0053] in, This indicates the average blood perfusion in the cardiac region;

[0054] The coefficient of variation of blood perfusion is expressed as the standard deviation of blood perfusion divided by the mean in this invention.

[0055] Indicates peak blood perfusion;

[0056] This represents the rate of change in blood perfusion relative to healthy individuals; the calculation method is as follows:

[0057]

[0058] in, This represents the reference value for the average blood perfusion in healthy individuals.

[0059] 3. Morphological feature acquisition and quantification

[0060] Data Acquisition Equipment: A 20-megapixel high-definition camera (Canon EOS R5) was used, along with a ring light, to capture close-up images of the "heart area" of both ears against a white background (300dpi resolution, JPEG format).

[0061] Quantitative scoring system: ;

[0062] in, Indicates morphological feature score,

[0063] In this embodiment, The "Auricular Acupoint Morphology Positive Scoring Scale" is used to score from 4 dimensions (0-2 points for each item, 0-8 points in total).

[0064] Table 1 Positive Scoring Table for Ear Acupoint Morphology

[0065] Rating Dimensions 0 points (negative) 1 point (mildly positive) 2 points (severely positive) papules / nodules none Diameter < 1mm, Quantity ≤ 2 Diameter ≥ 1mm, Quantity > 2 Vascular abnormalities No obvious blood vessels Slight dilation / torsion of blood vessels Significantly dilated, congested, or deformed blood vessels pigmentation none Pale red / light brown, area <0.5cm² Dark red / black, area ≥ 0.5cm² Skin depressions / bulges Smooth skin Slight dent / bulge (<0.5mm) Obvious depression / protrusion (≥0.5mm)

[0066] (ii) Data preprocessing

[0067] The collected multimodal features of auricular acupoints were standardized to eliminate individual differences and environmental interference. Specific steps included:

[0068] 1. Outlier Removal: The "3σ principle" is used to remove extreme values ​​(such as temperature differences). >3℃ or <-1℃, blood perfusion (>400PU or <50PU), ensure data validity;

[0069] 2. Standardization: Temperature and blood flow characteristics were standardized using Z-score to eliminate differences in baseline physiological state among different subjects; morphological scores were normalized using Min-Max, mapping the total score to the [0,1] interval;

[0070] 3. Missing value handling: If a sample is missing a single feature (such as blood flow data acquisition failure), the mean of the "same sex-same age group" will be used to fill the missing value; samples with ≥2 missing features will be directly removed to ensure data integrity.

[0071] (III) Characteristic screening and establishment of positive reaction criteria

[0072] Multidimensional statistical methods were used to screen core features strongly associated with the risk of coronary heart disease, and a positive reactant combination standard was established:

[0073] 1. Correlation Analysis: Spearman correlation analysis was used to calculate the correlation coefficient between the characteristics of each auricular acupoint and the gold standard for diagnosing coronary heart disease (Gensini score for the degree of coronary angiography stenosis). Features with r ≥ 0.4 were selected (preliminarily retained). , , , (A total of 4 core features)

[0074] 2. ROC Curve Cutoff Value Determination: For the initially retained features, plot the ROC curve, calculate the Youden index (sensitivity + specificity - 1), and determine the optimal cutoff value for each feature. ≥0.5℃ (Yorden Index 0.62);

[0075] ≤120PU (Yorden Index 0.58);

[0076] ≤-20% (Yorden Index 0.55);

[0077] ≥4 points (Youden Index 0.60);

[0078] 3. Positive reaction combination criteria: Samples that meet the above criteria of ≥2 characteristic cutoff values ​​are judged as "ear acupoint multimodal positive" and included in the high-risk candidate group.

[0079] (iv) Machine learning model construction and validation

[0080] A fusion early warning model combining "multimodal features of auricular acupoints + traditional risk factors" was constructed, and multiple algorithms were compared and optimized. The specific process is as follows:

[0081] 1. Input feature set:

[0082] Multimodal characteristics of auricular acupoints: (After standardization) (After standardization) (After standardization) (After normalization);

[0083] Traditional risk factors: age, sex (male = 1, female = 0), low-density lipoprotein cholesterol (LDL-C), high-sensitivity C-reactive protein (hs-CRP), smoking history (yes = 1, no = 0);

[0084] There are a total of 9 input features.

[0085] 2. Dataset partitioning and labeling:

[0086] Data source: 300 patients diagnosed with coronary heart disease (coronary angiography stenosis ≥50%), 200 high-risk patients with coronary heart disease (with ≥2 risk factors, stenosis 10%-50%), and 400 healthy controls (without risk factors, stenosis <10%), for a total of 900 samples;

[0087] Labeling criteria: The label is "whether or not the patient has coronary artery disease (stenosis ≥50%)" (patient = 1, no patient = 0);

[0088] Dataset partitioning: The dataset was divided into a training set (630 cases) and a test set (270 cases) in a 7:3 ratio. The training set was sampled using Bootstrap to generate 10 independent subsets for model ensemble training.

[0089] 3. Model Algorithm Selection and Optimization:

[0090] Candidate Algorithms: Random Forest (RF), Support Vector Machine (SVM), and 50-layer Residual Convolutional Neural Network (ResNet-50) were compared, and parameters were optimized through 5-fold cross-validation;

[0091] Core parameter optimization:

[0092] Random Forest: Number of decision trees = 500, maximum number of features = 5, minimum number of sample splits = 10;

[0093] Support Vector Machine: The kernel function is a radial basis function (RBF), with a penalty coefficient C=10 and gamma=0.1;

[0094] ResNet-50: Learning rate = 0.001, number of iterations = 100, dropout ratio = 0.3;

[0095] Imbalanced sample handling: A weighted cross-entropy loss function is introduced to address the issue of a low proportion of diseased samples in the training set. The formula is:

[0096]

[0097] in, Total loss;

[0098] n is the total number of samples;

[0099] i is the sample index;

[0100] Let be the label value of the i-th sample;

[0101] Let be the predicted value for the i-th sample;

[0102] The weighting factor is defined as:

[0103]

[0104] in Indicates the number of samples without disease;

[0105] Indicating diseased samples

[0106] 4. Model Training and Performance Validation:

[0107] Training process: Three algorithm models are trained separately using training set subsets, and the error is calculated using the cross-entropy loss function. The parameters are updated by backpropagation. The output results of the three models are fused through ensemble learning (voting method) to improve generalization ability.

[0108] Performance evaluation: Calculate core metrics on the test set:

[0109] Evaluation indicators Random Forest Support Vector Machine ResNet-50 ensemble model Sensitivity 0.82 0.78 0.85 0.87 Specificity 0.80 0.83 0.79 0.85 AUC (Area Under the ROC Curve) 0.85 0.83 0.86 0.89

[0110] Incremental value verification: The DeLong test was used to compare the AUC difference between the "traditional risk factor model only" and the "traditional risk factor + auricular acupoint multimodal feature model". The results showed that the AUC of the latter was significantly improved (0.76→0.89, p<0.001), which proved the incremental predictive value of auricular acupoint features.

[0111] 5. Validation of forward-looking early warning:

[0112] Study design: A 5-year follow-up study was conducted on 500 high-risk individuals with coronary artery disease (without baseline coronary artery disease) to record endpoint events (diagnosis of coronary artery disease or major adverse cardiovascular events, MACE).

[0113] Validation results: Kaplan-Meier curve analysis showed that the model predicted a significantly higher 5-year coronary heart disease incidence rate in the positive group (28.6%) than in the negative group (8.2%), with a hazard ratio (HR) of 3.2 (95% CI: 2.1-4.8); the net reclassification index (NRI) was 0.21, demonstrating that the model can effectively improve risk stratification.

[0114] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for constructing an early warning model for coronary heart disease based on multimodal features of auricular acupoints, characterized in that, Includes the following steps: Using ear region temperature characteristics, ear region blood flow characteristics, ear region morphological characteristics, and traditional risk factors as inputs to a machine learning model, and using whether or not someone has coronary heart disease as a label, the machine learning model is trained to obtain an early warning model for coronary heart disease.

2. The method for constructing an early warning model for coronary heart disease based on multimodal features of auricular acupoints according to claim 1, characterized in that, The traditional risk factors are age, sex, low-density lipoprotein cholesterol, high-sensitivity C-reactive protein, and smoking history.

3. The method for constructing an early warning model for coronary heart disease based on multimodal features of auricular acupoints according to claim 1, characterized in that, The ear region temperature characteristics are as follows: The blood flow characteristics of the ear region are as follows: The morphological features of the ear region ; in, This indicates the temperature difference between the cardiac region and the concha cymbidium region; This indicates the average blood perfusion in the cardiac region; This represents the rate of change in blood perfusion relative to healthy individuals; the calculation method is as follows: in, This represents the reference value for average blood perfusion in healthy individuals. The score represents the evaluation of the ear region morphology from four dimensions: papules / nodules, vascular abnormalities, pigmentation, and skin depressions / protrusions.

4. The method for constructing an early warning model for coronary heart disease based on multimodal features of auricular acupoints according to claim 1, characterized in that, The presence of coronary artery disease is labeled using the Gensini score, which indicates the degree of stenosis on coronary angiography.

5. The method for constructing an early warning model for coronary heart disease based on multimodal features of auricular acupoints according to claim 1, characterized in that, The machine learning algorithms include one or more ensemble models of random forest, support vector machine, and ResNet-50.

6. The method for constructing an early warning model for coronary heart disease based on multimodal features of auricular acupoints according to claim 5, characterized in that, The hyperparameters of the random forest are: number of decision trees = 500, maximum number of features = 5, minimum number of sample splits = 10; the hyperparameters of the support vector machine are: kernel function is radial basis kernel, penalty coefficient C = 10, gamma = 0.1; the hyperparameters of the ResNet-50 are: learning rate = 0.001, number of iterations = 100, dropout ratio = 0.

3.

7. The method for constructing an early warning model for coronary heart disease based on multimodal features of auricular acupoints according to claim 1, characterized in that, The ear region temperature characteristics were acquired using infrared thermal imaging, and the ear region blood flow characteristics were acquired using laser speckle imaging.

8. A method for early warning of coronary heart disease based on multimodal features of auricular acupoints, characterized in that, Ear temperature characteristics, ear blood flow characteristics, ear morphological characteristics, and traditional risk factors are used as inputs to the early warning model for coronary heart disease to obtain a coronary heart disease early warning risk score; the early warning model for coronary heart disease is the early warning model for coronary heart disease as described in any one of claims 1 to 7.

9. An electronic device system, characterized in that, include: Memory, used to store computer programs; A processor, when executing a program stored in memory, implements the steps of the method according to any one of claims 1 to 8.

10. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the method described in any one of claims 1 to 8.

Images