Method, system, device and medium for screening depression based on fundus color photography

Abstract

The present invention discloses a method, system, device, and medium for screening for depression based on fundus color photography. The method comprises the following steps: collecting a dataset of fundus color photography images of healthy individuals and patients with depression, measuring features of the image dataset, constructing a model based on a random forest algorithm, classifying the image data, validating and evaluating the model, and applying the model to a clinical workflow. The present invention uses fundus color photography images, depression diagnosis, and clinical risk factors to train and validate a model on confirmed patients with depression and healthy controls of the same age and gender. The trained random forest model is then used to distinguish between healthy individuals and patients with depression, providing a method and tool for early screening and identification of depression.

Description

Technical Field

[0001] The present invention relates to the field of medical disease screening, and more particularly to a method, system, device and medium for screening depression based on fundus color photography. Background Art

[0002] Depression is a common mental disorder characterized by prolonged low mood or loss of pleasure or interest in activities, affecting all aspects of life. Depression affects a large and diverse population. According to statistics, approximately 3.8% of the world's population suffers from depression, including 5% of adults (4% of men and 6% of women) and 5.7% of adults over 60 years old. The incidence of depression in women is approximately 50% higher than in men. Over 10% of pregnant women and women who have just given birth suffer from depression. Over 700,000 people commit suicide each year, making suicide the fourth leading cause of death among people aged 15-29. Therefore, the prevention, diagnosis, and treatment of depression are crucial. Current diagnostic methods for depression rely primarily on questionnaires and physicians' assessment of symptoms. These methods are often subjective, costly, and difficult to access, leading to missed or misdiagnosed cases. Therefore, there is an urgent need to develop objective, efficient, and convenient tools to assist in the assessment of depression.

[0003] Fundus color photography offers the unique advantages of being non-invasive, objective, efficient, and convenient, making it widely used for diagnosis and monitoring in ophthalmology. Detection of ocular diseases primarily relies on the evaluation of fundus color photography by professional physicians, a method that relies heavily on their extensive experience and expertise. With the rapid development of artificial intelligence (AI), particularly deep learning, AI models based on fundus color photography have demonstrated promising performance in the assessment of common ophthalmic diseases such as diabetic retinopathy, age-related macular degeneration, and glaucoma, demonstrating the enormous potential of fundus color photography AI. However, the current application of fundus color photography is limited to the diagnosis of ocular diseases and is rarely used for systemic diseases, which can also present diverse ocular manifestations, such as depression, hypertension, kidney disease, and diabetes. Numerous studies have confirmed that patients with depression exhibit varying degrees of fundus structural changes, such as alterations in vascular fractal dimension, optic disc structure, retinal microcirculation, and choroidal vascular density. Furthermore, as a non-invasive, low-cost, and easy-to-use assessment method, fundus color photography may provide a new avenue for early screening of depression.

[0004] However, for diseases such as depression that are not directly reflected in fundus images, current methods have the disadvantage that their accuracy and stability still need to be improved. These problems may lead to unfavorable factors such as insufficient sensitivity and accuracy, which limit the widespread application of fundus color photography in depression screening. Existing methods are too limited in the scope of use of fundus feature parameters, mainly focusing on the diagnosis of eye diseases, and are relatively lacking in the classification of healthy people and depression. In addition, the lack of feature extraction makes it impossible to fully obtain the features related to depression in fundus images, resulting in difficulties in accurately distinguishing healthy people from depression. This proposal aims to establish a classification model for healthy people and depression based on fundus color photography, which can efficiently and accurately screen out patients with depression from the population. Summary of the Invention

[0005] The present invention aims to overcome at least one defect in the prior art and provide a method, system, device and medium for screening depression based on fundus color photography, which is used to solve the problems of low accuracy and instability in distinguishing healthy people from depression patients through fundus images.

[0006] The technical solution adopted by the present invention is a method for screening depression based on fundus color photography, comprising the following steps:

[0007] Step S100: collecting an image dataset containing healthy individuals and depression patients;

[0008] Step S200: performing feature measurement on the image data set;

[0009] Step S300: Building a model based on random forest to classify image data;

[0010] Step S400: verifying and evaluating the model;

[0011] Step S500: Apply the model to the clinical workflow to provide real-time disease screening.

[0012] First, we collected a large number of fundus color images from the UK Biobank, community organizations, and hospitals. We also collected information on depression diagnoses (including whether patients were diagnosed with depression, whether they had a preliminary diagnosis of eye disease, and whether they had other systemic diseases, such as cardiovascular and cerebrovascular disease, neurodegenerative disease, and psychiatric illness, as well as the specific diagnosis) and clinical risk factors (gender, age, and body mass index). We then trained and validated the model using patients diagnosed with depression and healthy controls of the same age and gender.

[0013] In order to increase the extraction of depression-related features, special measurement indicators are needed. Therefore, the feature measurements of the image dataset are designed to include the following indicators:

[0014] Optic disc tilt angle: the angle between the major axis of the smallest outer ellipse and the horizontal line;

[0015] Optic disc roundness:

[0016] in, C represents the roundness of the optic disc, F Represents the optic disc area, that is, the number of pixels occupied by the optic disc area, R Represents the minimum circumferential radius of the optic disc.

[0017] To enable early screening and identification of depression, a model was constructed to distinguish healthy people from patients with depression based on fundus images. The random forest model was constructed based on measured fundus image features and clinical risk factors, and included the following steps:

[0018] Step S310: Randomly select samples and use the bootstrap method with replacement to select samples from the original dataset containing fundus image features, depression diagnosis and clinical risk factors. D Randomly draw samples from

[0019] Step S320: randomly selecting features, including optic disc tilt angle, optic disc roundness, gender, age, and body mass index, and randomly selecting features for splitting at each node of each decision tree;

[0020] Step S330: Construct a decision tree using the CART (Classification and Regression Trees) algorithm, and use the Gini Index as one of the criteria for selecting features and dividing nodes.

[0021] Step S340: training the random forest model by using the preprocessed and feature-selected data.

[0022] During training, the model learns how to distinguish healthy individuals from those with depression based on input fundus color image features and clinical risk factors. The model also simplifies the required data, using only fundus color images, gender, age, and body mass index to identify depression. This makes it more suitable for use in areas with limited medical resources and increases the accessibility of depression screening.

[0023] When building a decision tree, the Gini index is used as one of the criteria for selecting features and dividing nodes. The Gini index measures the probability that two randomly selected samples from a dataset will have inconsistent class labels. A smaller Gini index indicates a higher dataset purity, meaning that the samples are more likely to belong to the same class.

[0024] The calculation of the Gini index is K categories, hereK 2, including data sets of healthy people and patients with depression D , and its Gini index is defined as:

[0025]

[0026] in, p k It is a dataset D Middle k The proportion of class samples;

[0027] feature A Including: optic disc tilt angle, optic disc roundness, gender, age and body mass index, for characteristics A A division under D 1 and D 2. Dataset D According to the characteristics A Divided into two parts, the Gini index after division is:

[0028]

[0029] Among them, D ∣、∣ D 1|and| D 2| represent data sets respectively D 、 D 1 and D The sample size is 2.

[0030] The specific process of building a decision tree based on the CART algorithm includes the following steps:

[0031] Step S321: Calculate the Gini index of each feature in the data set, select the feature with the smallest Gini index as the optimal partitioning feature, and divide the data set into different subsets according to different values ​​of the feature;

[0032] Step S322: Repeat step S321 for each subset until a stop condition is met;

[0033] Step S333: Pruning the decision tree.

[0034] The stopping conditions of step S322 include that the number of samples in the subset is less than the minimum number of samples of the algorithm parameters, all samples belong to the same category, there are no more features to be divided, etc. In step S333, in order to avoid overfitting, the decision tree is pruned to remove leaf nodes with low gain and simplify the model.

[0035] To ensure the robustness and generalization ability of the model, the random forest model used five-fold cross-validation for internal validation and fundus image data and depression diagnosis for external validation. The fundus image data used relevant data from 300 subjects in hospitals and communities whose personal identity information was concealed to provide sufficient external validation samples for model training.

[0036] In order to more intuitively show the difference between the model's predicted results and the actual results, the evaluation uses cross entropy to evaluate the difference between the probability distribution predicted by the model and the actual distribution:

[0037]

[0038] in, x represents a possible value of a random variable (1 or 0, 1 represents depression and 0 represents healthy control), n is the number of possible values ​​of the random variable, p ( x )and q ( x ) respectively represent x In distribution p and q The probability in the formula specifically indicates that in the real distribution P Next, use the predictive distribution Q The average encoding length during encoding, that is, the average information entropy, is expressed in bits or nats. The smaller the cross entropy value, the closer the two probability distributions are. The cross entropy here is a small value such as 0.2231. The lower the better.

[0039] A depression screening system based on fundus color photography, comprising:

[0040] An image collection module is used to collect a fundus color image dataset containing healthy individuals and patients with depression;

[0041] A measurement module, configured to: perform feature measurement on the image data set;

[0042] Generate a module to: build a random forest model to classify image data;

[0043] Verification and evaluation module: verify and evaluate the random forest model;

[0044] The system has the ability to process and analyze fundus color photographs in real time and provide risk assessments for the diseases, and also includes a user-friendly interface to improve the efficiency and accuracy of clinical workflow.

[0045] The present application also provides a computer device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is executed by the processor, any one of the above methods for screening depression based on fundus color photography is implemented.

[0046] The present application also provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, any of the above methods for screening depression based on fundus color photography is implemented.

[0047] Compared with the prior art, the present invention has the following advantages and technical achievements:

[0048] (1) Provide early screening for depression: By measuring the features of fundus images, the extraction of features related to depression is increased. The random forest model is constructed based on the measured fundus image features and clinical risk factors. It can distinguish between healthy people and patients with depression, providing an objective and convenient tool for early screening and identification of depression.

[0049] (2) Improvement in popularity: By simplifying the data and using fundus color photographs and gender, age, and body mass index to identify depression, the system is more suitable for use in areas with scarce medical resources, significantly improving the popularity of depression screening.

[0050] (3) The model has strong robustness and generalization ability: The fundus image dataset was established using 400 healthy controls of the same age and gender with a clear diagnosis of depression in a public database (personal identity information was concealed) for model training and verification, including fundus photo datasets, depression diagnosis and clinical risk factors (gender, age, body mass index). In addition, relevant data from 300 subjects in hospitals and communities (personal identity information was concealed) were collected and used to provide sufficient external verification samples for model training.

[0051] (4) Economic and social benefits: This invention is expected to significantly reduce medical costs due to misdiagnosis and missed diagnosis, and may reduce the overall treatment costs of depression through early screening. In addition, objective, efficient and convenient classification tools will improve the accessibility of medical services and show potential in improving the level of public health. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] The accompanying drawings are for illustrative purposes only and are not to be construed as limiting the present invention. To more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly describes the accompanying drawings required for describing the embodiments. Obviously, the accompanying drawings described below represent only some embodiments of the present invention. Persons skilled in the art can, without inventive effort, derive other drawings from these drawings.

[0053] Figure 1 Flow chart of the method of the present invention.

[0054] Figure 2 This is a diagram of the depression screening model based on fundus color photography of the present invention.

[0055] Figure 3 This is a diagram of the depression screening model based on fundus color photography and clinical risk factors of the present invention.

[0056] Figure 4 This is a flow chart of the random forest model of the present invention.

[0057] Figure 5 This is the fundus color photograph discrimination diagram for healthy and depressed individuals of the present invention. DETAILED DESCRIPTION

[0058] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts shall fall within the scope of protection of the present invention.

[0059] like Figure 1 As shown, the present invention provides a method for screening depression based on fundus color photography, comprising the following steps:

[0060] Step S100: collecting an image dataset containing healthy individuals and depression patients;

[0061] Step S200: performing feature measurement on the image data set;

[0062] Step S300: Building a model based on random forest to classify image data;

[0063] Step S400: verify and evaluate the model.

[0064] Step S500: Apply the model to the clinical workflow to provide real-time disease screening.

[0065] The image dataset includes a large number of fundus color images collected from UK Biobank, communities, and hospitals, depression diagnosis (mainly including whether depression is diagnosed, whether there is a preliminary diagnosis of eye diseases and other systemic diseases such as cardiovascular and cerebrovascular diseases, neurodegenerative diseases, and mental illnesses, as well as the specific diagnosis) and clinical risk factors (gender, age, and body mass index). Feature measurements are performed on fundus images of healthy individuals and patients with depression, and features related to depression are extracted. A model is constructed based on the random forest model, image data is classified, and the model is verified and evaluated to improve the model, providing an objective and convenient tool for the early screening and identification of depression.

[0066] In this embodiment, if Figure 2 As shown in the figure, the input fundus image is subjected to feature measurements of the optic disc tilt angle and the optic disc roundness, and the input model is input with features such as gender, age and body mass index to distinguish between healthy people and people with depression.

[0067] The characteristic measurements of the optic disc tilt angle and optic disc roundness are specifically as follows:

[0068] The optic disc tilt angle is the angle between the major axis of the smallest outer ellipse and the horizontal line.

[0069] Optic disc roundness:

[0070] in, C represents the roundness of the optic disc, F Represents the optic disc area, that is, the number of pixels occupied by the optic disc area, R Represents the minimum circumferential radius of the optic disc.

[0071] By measuring the features of optic disc tilt angle and optic disc roundness, the extraction of depression-related features is increased, thereby improving the accuracy of the model.

[0072] In this embodiment, combined with Figure 3 and Figure 4 As shown in the figure, the model used in this method is a random forest model. The model is constructed based on the measured fundus image features and clinical risk factors. It can distinguish between healthy people and patients with depression and is suitable for early screening and identification of depression. The features of the optic disc tilt angle and optic disc roundness of the fundus image, gender, body mass index and age are input into the random forest model. The random forest model is used to train the sample set. D Randomly select samples and use the bootstrap sampling method with replacement (Bootstrap) from the training sample set D Randomly select samples from different sample sets D 1, D 2, D3, etc. For different sample sets, features such as optic disc tilt angle, optic disc roundness, gender, age, and body mass index were randomly selected and input into a decision tree. At each node in the decision tree, a feature was randomly selected for splitting. The Gini index was used as the criterion for selecting features and splitting nodes. A random forest model was trained using the preprocessed and feature-selected data.

[0073] The Gini index measures the probability that two samples randomly selected from a data set have inconsistent category labels. The smaller the Gini index, the higher the purity of the data set, that is, the greater the possibility that the samples belong to the same category. The calculation of the Gini index is as follows: K categories, here K 2, including data sets of healthy people and patients with depression D , and its Gini index is defined as:

[0074]

[0075] in, p k It is a dataset D Middle k The proportion of class samples;

[0076] feature A Including: optic disc tilt angle, optic disc roundness, gender, age and body mass index, for characteristics A A division under D 1 and D 2. Dataset D According to the characteristics A Divided into two parts, the Gini index after division is:

[0077]

[0078] Among them, D ∣、∣ D 1|and| D 2|represents the data sets respectively D 、 D 1 and D The sample size is 2.

[0079] The decision tree in the random forest model described above is constructed based on the CART algorithm. Specifically, the Gini index of each feature in the dataset is calculated, the feature with the smallest Gini index is selected as the optimal partitioning feature, and the dataset is divided into different subsets based on the different values ​​of the feature. The Gini index of each feature in the dataset is repeatedly calculated for each subset, the feature with the smallest Gini index is selected as the optimal partitioning feature, and the dataset is divided into different subsets based on the different values ​​of the feature until the stopping condition is met (the number of samples in the subset is less than the minimum sample size of the algorithm parameter, all samples belong to the same category, there are no more features to partition, etc.). To avoid overfitting, the decision tree is pruned to remove leaf nodes with low gain and simplify the model. To ensure the robustness and generalization ability of the model, the random forest model was internally validated using five-fold cross-validation and externally validated using fundus image data and depression diagnosis. The fundus image data used was related data from 300 subjects in hospitals and community settings, with their personal identification information anonymized, providing sufficient external validation samples for model training.

[0080] In this embodiment, if Figure 5 As shown, it reflects the difference between the probability distribution predicted by the model and the true distribution. The cross entropy is used for evaluation, specifically:

[0081]

[0082] in, x represents a possible value of a random variable (1 or 0, 1 represents depression and 0 represents healthy control), n is the number of possible values ​​of the random variable, p ( x )and q ( x ) respectively represent x In distribution p and q The probability of

[0083] The formula specifically shows that in the real distribution P Next, use the predictive distribution Q The average encoding length during encoding, that is, the average information entropy, is expressed in bits or nats. The smaller the cross entropy value, the closer the two probability distributions are. The cross entropy here is a small value such as 0.2231. The lower the better.

[0084] In addition, an embodiment of the present application also provides a system for screening depression based on fundus color photography, including an image collection module for collecting fundus color photography image datasets of healthy individuals and patients with depression, a measurement module for measuring features of the image dataset, a generation model for constructing a random forest model and classifying image data, and a verification and evaluation module for verifying and evaluating the random forest model. Using fundus color photography images, the diagnosis of depression, and clinical risk factors, a model is trained and verified on confirmed patients with depression and healthy controls of the same age and gender to obtain a trained model that is used to distinguish between healthy people and patients with depression, providing a convenient tool for early screening and identification of depression. The system also has the ability to process and analyze fundus color photography in real time and provide risk assessments for the disease. The system also includes a user-friendly interface to improve the efficiency and accuracy of clinical workflows.

[0085] In addition, an embodiment of the present application also provides a computer device, including: a memory, a processor, and a computer program stored in the memory and runnable on the processor. When the computer program is executed by the processor, the above-mentioned method for screening depression based on fundus color photography is implemented.

[0086] In addition, an embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, any of the above methods for screening depression based on fundus color photography is implemented.

[0087] The preferred embodiments of the present invention disclosed above are intended only to help illustrate the present invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the present invention to specific embodiments. Obviously, many modifications and variations are possible based on the contents of this specification. These embodiments are selected and described in detail herein to better explain the principles and practical applications of the present invention, thereby enabling those skilled in the art to better understand and utilize the present invention. The present invention is limited only by the claims and their full scope and equivalents.

Claims

1. A method for screening depression based on fundus color photography, characterized in that: The following steps are involved: Step S100: collecting an image dataset containing healthy individuals and depression patients; Step S200: measuring the optic disc tilt angle and optic disc roundness of the image data set; Step S300: constructing a random forest model based on fundus image features, gender, body mass index, and age factors to classify the image data; Step S400: verifying and evaluating the model; Step S500: applying the model to clinical workflow; The optic disc tilt angle and optic disc roundness are specifically: Optic disc tilt angle: the angle between the major axis of the smallest outer ellipse and the horizontal line; Optic disc roundness: Where C represents the circularity of the optic disc, F represents the optic disc area, that is, the number of pixels occupied by the optic disc area, and R represents the minimum circumference radius of the optic disc; The random forest model is constructed based on fundus image features, gender, body mass index and age factors, including the following steps: Step S310: randomly selecting samples, using the bootstrap sampling method with replacement, randomly extracting samples from the original dataset D containing fundus image features and depression diagnosis, and constructing multiple decision trees; Step S320: randomly selecting features, including optic disc tilt angle, optic disc roundness, gender, age, and body mass index, and randomly selecting features for splitting at each node of each decision tree; Step S330: Construct a decision tree using the CART (Classification and Regression Trees) algorithm, and use the Gini Index as one of the criteria for selecting features and dividing nodes. Step S340: training the random forest model by using the preprocessed and feature-selected data; The smaller the Gini index is, the higher the purity of the data set is, that is, the greater the possibility that the samples belong to the same category; The calculation of the Gini index, for a dataset D containing K categories, where K is 2, including healthy people and patients with depression, its Gini index is defined as: Among them, pk is the proportion of samples of the kth class in the data set D; Feature A includes: optic disc tilt angle, optic disc roundness, gender, age, and body mass index. For a partition D1 and D2 under feature A, that is, the data set D is divided into two parts according to feature A, the Gini index after the partition is: Among them, |D|, |D1|, and |D2| represent the number of samples in datasets D, D1, and D2, respectively; The specific process of constructing a decision tree based on the CART algorithm includes the following steps: Step S321: Calculate the Gini index of each feature in the data set, select the feature with the smallest Gini index as the optimal partitioning feature, and divide the data set into different subsets according to different values ​​of the feature; Step S322: Repeat step S321 for each subset until a stop condition is met; Step S333: Pruning the decision tree.

2. The method for screening depression based on fundus color photography according to claim 1, characterized in that: The random forest model was internally validated using five-fold cross validation and externally validated using fundus image data and depression diagnosis.

3. The method for screening depression based on fundus color photography according to claim 2, characterized in that: The evaluation uses cross entropy to evaluate the difference between the probability distribution predicted by the model and the true distribution: Where x represents a possible value of the random variable, which can be 1 or 0, with 1 representing depression and 0 representing a healthy control. n is the number of all possible values ​​of the random variable. p(x) and q(x) represent the probabilities of x in distributions p and q, respectively. The formula specifically represents the average coding length when encoding using the predicted distribution Q under the true distribution P, that is, the average information entropy, in bits or nats. The smaller the cross entropy value, the closer the two probability distributions are.

4. A system for screening depression based on fundus color photography constructed according to the method for screening depression based on fundus color photography according to any one of claims 1 to 3, characterized in that: include: An image collection module is used to collect a fundus color image dataset containing healthy individuals and patients with depression; A measurement module, configured to: measure the optic disc tilt angle and optic disc roundness of the image data set; A generation module is used to: construct a random forest model based on fundus image features, gender, body mass index, and age factors to classify image data; Verification and evaluation module: verify and evaluate the random forest model; The depression screening system based on fundus color photography has the ability to process and analyze fundus color photography in real time and provide risk assessment of the disease.

5. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the method for screening depression based on fundus color photography according to any one of claims 1 to 4 is implemented.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for screening depression based on fundus color photography according to any one of claims 1 to 4 is implemented.

Images