Human hair detection method for androgenetic alopecia, digital diagnosis and treatment method, system, equipment and medium

By constructing a hair state detection neural network model with adaptive cropping and multi-grained feature extraction, combining hair images and specialized disease scale data, the decision tree is generated, and the problem of lack of personalized and dynamic monitoring of existing hair loss treatment methods is solved, and the problem of the lack of personalized and dynamic monitoring of existing hair loss treatment methods is realized. Efficient personalized diagnosis and treatment of androgenic hair loss.

CN118982846AActive Publication Date: 2024-11-19WEST CHINA HOSPITAL SICHUAN UNIV

Patent Information

Application Number
CN202411132437.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-19
Publication Date
2024-11-19
Estimated Expiration
2044-08-19

AI Technical Summary

Technical Problem

The existing hair loss treatment methods lack personalized and dynamic monitoring functions, the drug treatment cycle is long, the onset of effect is slow, the hair follicle transplantation costs are high and the recovery period is long. It is difficult for the existing hair health testing system to quickly give preliminary diagnostic opinions, and there is external component dependence. , causing inconvenience to users.

Method used

A hair state detection neural network model based on adaptive cropping and multi-grained feature extraction is constructed, and a decision tree is generated by combining hair images and specialized disease scale questionnaire data, providing personalized diagnosis and treatment opinions and intervention suggestions.

Benefits of technology

It improves the accuracy of hair image detection and the auxiliary diagnosis ability of digital diagnosis and treatment, outputs system medical advice or intervention opinions, and realizes personalized and dynamic monitoring of androgenic hair loss.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118982846A_ABST
    Figure CN118982846A_ABST
Patent Text Reader

Abstract

The invention discloses a human hair detection method, a digital diagnosis and treatment method, a digital diagnosis and treatment system, equipment and a medium for androgenetic alopecia, is applied to the technical field of artificial intelligence medicine, and aims to solve the problem that in the prior art, only the health state of hair can be detected, and preliminary diagnosis and treatment opinions cannot be given. According to the invention, a hair state detection neural network model based on adaptive cutting and multi-granularity feature extraction is constructed, and accurate detection is carried out on a hair image; and combining the hair state detection result with the quantitative data of the special disease scale questionnaire to generate a decision tree model about hair classification, thereby outputting system doctor seeing opinions or intervention opinions.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the field of artificial intelligence medical technology, and in particular relates to a hair loss detection technology, diagnosis and treatment technology and system. Background Art

[0002] Androgenic alopecia is a common hair disease that affects a wide range of men and women. Although there are currently a variety of treatments for androgenic alopecia, such as drug therapy, lifestyle adjustments, and hair follicle transplantation, each of these methods has its own limitations. For example, drug treatment has a long cycle and slow onset, which often leads to low patient compliance, that is, when patients receive medical intervention, they are less likely to follow the advice or instructions of medical professionals to implement the treatment plan; lifestyle adjustments also lack effective prompts and supervision, and it is difficult to achieve a good treatment effect; hair follicle transplantation surgery is expensive, has a long recovery period, and has certain requirements for the patient's baldness type. Existing hair loss treatment methods lack personalization and dynamic monitoring functions. With the development of information technology, some hair health assessment tools based on mobile applications have emerged, but these tools usually only provide basic health advice and lack in-depth medical evaluation and personalized intervention plans.

[0003] In the prior art, hair health detection is mainly divided into two branches. One is to use a high-speed camera component to take pictures of the hair, and then an online doctor will make a judgment on the pictures. The other is to develop hardware equipment, which is worn in the detection area for a long time, and the optical sensor is used to track and detect the health of the hair. After the detection is completed, the tracking data stored in the device needs to be exported for further judgment. The above two methods only detect the health status of the hair and cannot quickly give a preliminary diagnosis. At the same time, this type of system also has a large number of external components to complete the detection, which brings great inconvenience to users. Summary of the invention

[0004] In order to solve the above technical problems, the present invention proposes a human hair digital diagnosis and treatment method, system, equipment and medium for androgenic alopecia.

[0005] One of the technical solutions adopted by the present invention is: a human hair detection method for androgenic alopecia, comprising:

[0006] A1. Construct a neural network model for hair state detection based on adaptive cropping and multi-granularity feature extraction;

[0007] A2. Obtain hair images of different states and offline medical records and clinical opinions corresponding to the hair images, and annotate each hair image according to the offline medical records and clinical opinions to obtain a training data set;

[0008] A3, using the training data set of step A2 to train the hair state detection neural network model constructed in step A1;

[0009] A4. Input the hair image to be detected into the trained hair state detection neural network model to obtain the hair state detection result.

[0010] The second technical solution adopted by the present invention is: a digital diagnosis and treatment method for human hair for androgenic alopecia, comprising:

[0011] B1. Data acquisition steps include:

[0012] B11. Using the human hair detection method for androgenic alopecia according to claim 1, obtaining the hair detection result of the hair image to be detected;

[0013] B12. Obtain the data of the disease-specific questionnaire, which includes: hair density, hair diameter, hair follicle health, genetic history, eating habits, living habits, mental stress level and medication status;

[0014] B2. Data preprocessing steps include:

[0015] B21. Use Cronbach's coefficient to quantify the data of the disease-specific scale questionnaire to obtain quantitative data;

[0016] B22, classifying the hair detection result of step B11 and the quantitative data obtained in step B21 according to BASP, marking them as feature selection criteria, and obtaining a feature data set;

[0017] B23. Calculate the information entropy of each feature in the feature data set, and calculate the information gain based on the information entropy;

[0018] B3. Data analysis step: information gain is used to provide a basis for decision tree generation for the C4.5 decision tree algorithm, thereby generating a diagnosis and treatment opinion and intervention opinion decision tree with different intervention levels for different types and severity of hair conditions.

[0019] The third technical solution adopted by the present invention is: a human hair digital diagnosis and treatment system for androgenic alopecia, comprising: a data processing module, and an information acquisition module electrically connected to the data processing module; the information acquisition module is used to obtain hair images and special disease scale questionnaire information; the data processing model outputs medical advice or intervention advice based on the hair images and special disease scale questionnaire information.

[0020] The fourth technical solution adopted by the present invention is: a computer device, comprising: a memory, a processor and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of a digital diagnosis and treatment method for human hair for androgenic alopecia.

[0021] The fifth technical solution adopted by the present invention is: a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of a digital diagnosis and treatment method for human hair for androgenic alopecia.

[0022] Beneficial effects of the present invention: The present invention constructs a neural network model for hair status detection based on adaptive cropping and multi-granularity feature extraction. Compared with the prior art, it can more efficiently crop useless areas in hair images, improve the ability to extract features of different granularities, and ultimately effectively improve the detection accuracy of hair images; and combines hair status data with quantitative data from a disease-specific scale questionnaire to generate a decision tree for hair classification. Compared with the prior art, in the process of digital hair diagnosis and treatment, the hair status based on image prediction and the disease-specific scale questionnaire in text form can be comprehensively considered. The content of the questionnaire includes hair density, hair diameter, hair follicle health, genetic history, eating habits, living habits, mental stress level, and medication status, etc., thereby improving the auxiliary diagnosis capability of digital diagnosis and treatment, thereby outputting system consultation opinions or intervention opinions. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 A schematic diagram of a human hair health detection structure provided by an embodiment of the present invention;

[0024] Figure 2 A structural diagram of a human hair health detection network model provided by an embodiment of the present invention;

[0025] Figure 3 The human hair digital diagnosis and treatment system architecture provided by the embodiment of the present invention;

[0026] Figure 4 A schematic diagram of the functional modules of the human hair digital diagnosis and treatment system provided by an embodiment of the present invention;

[0027] Figure 5 A system block diagram provided for an embodiment of the present invention. DETAILED DESCRIPTION

[0028] To facilitate those skilled in the art to understand the technical content of the present invention, the present invention is further explained below with reference to the accompanying drawings.

[0029] like Figure 1 As shown, the present invention provides a human hair digital diagnosis and treatment method for androgenic alopecia, comprising the following steps:

[0030] Step 1: Prepare 1,000 hair image samples of different patients and 1,000 corresponding offline medical records and clinical opinions, as well as 1,000 questionnaires at three time points after the diagnosis of the image samples.

[0031] like Figure 2 The figure shows the structure diagram of the human hair health detection network model of the present invention. The detection network is based on adaptive cropping and multi-granularity feature extraction, using ResNet-50 as the backbone network, and adding an adaptive cropping module before the first residual block of ResNet-50 to achieve automatic cropping of the hair area. The adaptive cropping module consists of a 3*3 convolution layer, group normalization, ReLU activation layer and a self-attention module, which is mainly used to output hair area and position mask information. The ResNet-50 network mainly consists of 4 residual blocks. The original image is input into the adaptive cropping module. The image output by the adaptive cropping module is multiplied with the original image and then input into the first residual module of ResNet-50. The output of the first residual block is multiplied with the output of the adaptive cropping module as the input of the second residual block of ResNet-50. The output of the second residual block is multiplied with the output of the adaptive cropping module as the input of the third residual block of ResNet-50. The output of the third residual block is multiplied with the output of the adaptive cropping module as the input of the fourth residual block, thereby realizing the automatic cropping of the hair area. The 4 residual blocks in ResNet-50 output feature maps of 4 different sizes: 32*32, 64*64, 128*128, and 256*256 respectively. A multi-grain feature network is introduced after the ResNet-50 backbone network, and the features output by the four residual blocks are input into the four reshapes (reconstructed sizes) of the multi-grain feature network to achieve feature data conversion. The converted data is input into a fully connected layer to finally output the hair detection results, including average color, texture, morphology, edge length and contour, glossiness, etc.

[0032] After the detection model is constructed, 1,000 gross image data with different hair conditions are collected, and the hair condition type data and hair area data are annotated according to the collected offline medical information and clinical opinions. The collected data is divided into training set data and test set data in a ratio of 7:3 according to the number of data, and the Adam optimizer and learning rate of 0.08 are set for training. By dynamically setting the number of epochs and drawing the training and validation loss curves, the changes in training and validation losses are observed. When the detection model still has a continuous small decrease in the validation loss process in multiple epochs, it indicates that there is still room for improvement. The number of epochs can be adjusted appropriately until the validation loss between multiple iterations of the model is observed to be less than 0.0001, which means that the validation loss of the model is no longer improving, and the training of the detection model can be stopped. At this time, it is a maturely trained detection model.

[0033] The medical information includes: the specific conditions of the patient during offline medical treatment, such as the patient's low hair density and family history of the disease, the patient's living habits with a single diet, etc., and the doctor's diagnosis based on the patient's specific conditions, such as the offline diagnosis of hereditary hair loss (androgenic alopecia); the clinical opinion refers to the treatment plan corresponding to the patient's condition, such as the corresponding clinical opinion given in this embodiment is: adjust the diet structure and increase nutrition; take supplements rich in vitamins and minerals every day; maintain a regular schedule and reduce stress; and use drugs such as minoxidil to promote hair growth.

[0034] In this embodiment, when training the hair detection model, each input hair image is accompanied by the corresponding offline medical consultation information and corresponding text description of the clinical opinion, that is, the corresponding text description of the medical consultation information and the clinical opinion is used as the annotation of the hair image; when the hair image without offline medical consultation is input into the trained hair detection model, the hair detection result output by the model will be accompanied by the matching annotation, and the annotation is specifically the text description of the matching preliminary diagnosis opinion and clinical opinion.

[0035] The detection process includes the contents of steps 2-4;

[0036] Step 2: cropping the hair image to obtain the hair area and position mask information in the hair image;

[0037] Step 3: Input the hair area and position mask information in the hair image into ResNet-50 to obtain feature maps of different scales;

[0038] Step 4: Input the feature maps of different scales obtained in step 3 into the multi-grain feature network to obtain the hair detection result;

[0039] Step 5: Use Cronbach's (reliability) coefficient to quantify the results of the questionnaire survey and obtain the correlation strength between the questions in the questionnaire and the types in the BASP classification system;

[0040] The correlation strength between the questions in the questionnaire and the types in the BASP classification system is obtained. The Cronbach (reliability) coefficient formula is:

[0041]

[0042] Among them, α is the reliability coefficient, K is the number of test questions, represents the score variation of all subjects on question i, is the variance of the total scores of all subjects.

[0043] Step 6: After the relevant coefficients are obtained through the questionnaire survey, the hair detection data obtained in step 4 and the quantitative coefficients of the questionnaire survey obtained in step 5 are annotated as feature selection criteria according to the BASP classification, wherein the feature data includes the BASP classification and its level (F, M, C, U and its specific level), hair density, hair diameter, hair follicle health, genetic history, eating habits, living habits, mental stress level, treatment history, etc. The information entropy, feature entropy and information gain of the feature are calculated. The smaller the information entropy, the purer the data and the greater the information gain.

[0044] Select the data with the largest information gain and the most effective feature division in the same type as the basis for feature classification and some nodes in the decision tree. Specifically:

[0045] After obtaining the information entropy, feature entropy, and information gain data, the information gain rate is obtained through information gain and information entropy. The information gain rate provides a basis for decision tree generation for the C4.5 decision tree algorithm. A tree node is generated according to the data set of the selected features, and the recursion is repeated continuously to generate a diagnosis and treatment opinion and intervention opinion decision tree with different intervention levels for different types and severity of hair conditions.

[0046] During the decision tree generation process, the initial information entropy calculation formula is:

[0047]

[0048] where p i is the proportion of samples belonging to the (i)th category in the dataset.

[0049] The characteristic information entropy calculation formula is:

[0050]

[0051] where Values(A) represents the set of all possible values ​​of feature(A), Sv represents the number of samples in the data set when the feature (A) takes the value (V), (|S|) is the total number of samples, and H(S v ) indicates that the sub-dataset (S) is v ), the calculation method is the same as the initial information entropy; the value of the feature here is the quantization result of the relevant feature in step 5, that is, the value of α. The information gain calculation formula is:

[0052] Gain(S,A)=H(S)-H(S|A)

[0053] Similar calculations are performed on all features, and the features with the largest information gain are selected as the partitioning features and as some nodes in the decision tree. Through the partitioning of the decision tree, the system can output accurate diagnosis and treatment opinions and intervention opinions.

[0054] The final output of the decision tree is a text content composed of multiple condition matches. The conditions here specifically refer to various features, such as: average color, texture, shape, edge length and contour, glossiness, etc., including the detection indicators in the medical record and the result items of the questionnaire information survey; each feature is accompanied by a label, that is, the corresponding text description of the corresponding medical information and clinical opinions. The text descriptions corresponding to multiple features are combined and output, which is the preliminary diagnosis and treatment opinions and intervention opinions for the current hair image to be detected.

[0055] For example, according to the various features contained in the test results of a hair image to be tested, the matching diagnosis opinion is: hereditary hair loss, and the matching intervention suggestions are: adjust the diet structure and increase nutritional intake; take supplements rich in vitamins and minerals every day; maintain a regular schedule and reduce stress; minoxidil can be used to promote hair growth. The system will then output the matching diagnosis opinion and intervention opinion.

[0056] Step 1 also includes labeling the hair in the hair image according to the type of the BASP classification system to obtain the labeling data and mask position information. In actual application, the labeling data, mask position and image will be submitted to the system. After submission, the system will automatically jump to the corresponding disease scale questionnaire survey interface. In the interactive interface, the user's physical health and mental state in the past six months, medication in the past three months, average sleep time and sleep depth in the past month, etc. are answered in the questionnaire, and submitted to the system for storage. After the system obtains the questionnaire data submitted by the user, it uses the Cronbach (reliability) coefficient to quantify the questionnaire survey results.

[0057] The system outputs diagnosis and treatment opinions and intervention opinions. The system uploads the prepared diagnosis data and clinical opinions. The system generates a specific home care plan for the user through offline diagnosis data and clinical opinions, and updates the execution of the plan through specific methods. During the plan update process, the prepared follow-up questionnaire survey is uploaded at the planned time point. The system quantifies the questionnaire survey results and re-enters them into the detection model. Through regression training, the system improves the diagnosis and treatment opinions and intervention opinions of the user.

[0058] Users manage their personal information and related hair health issues on their mobile phones. Users take photos with their mobile phones to obtain hair images, and choose to upload the obtained hair images in the mobile phone APP. After uploading, the system will generate an electronic questionnaire for a special disease scale based on the user's status information, and trigger the mobile phone APP page to jump to the questionnaire filling page. The user fills in the information in the questionnaire according to his or her actual status and submits the completed questionnaire information to the application. The application system uses the Cronbach (reliability) coefficient to perform relevant calculations on the result data of the questionnaire information based on the submitted questionnaire information, and obtains the correlation strength corresponding to the type of questions in the questionnaire and the type in the BASP classification system. The obtained data and the uploaded hair images are passed into the trained detection model together. The detection model outputs the corresponding detection diagnosis opinions and intervention opinions through calculation, and transmits them to the application system for feedback to the user's mobile phone APP, and reminds the user to go for offline medical treatment according to the output opinions.

[0059] The user visits the doctor offline, obtains the medical records and clinical opinions, and opens the mobile phone APP to upload the medical records and clinical opinions to the system. The system passes the hair image submitted in step 2 and the data obtained from the offline visit to the detection model. The detection model calculates and outputs the user's home care plan for hair and transmits it to the application. The application feeds back the care plan to the user, generates user-specific home care plan tasks, and reminds the user of hair care messages in real time.

[0060] After the home care plan is generated, during the update process, the system will monitor the user's sleep duration, sleep quality, medication name and duration through the mobile phone APP, and issue a follow-up questionnaire on hair health status to the user at the planned time, and remind the user to fill in the real information. After the user submits the follow-up survey data, the system will pass the monitored sleep and medication related data and the quantitative data of the follow-up questionnaire survey results to the detection model. The detection model calculates the recent data and the original data and outputs the update of the family plan.

[0061] In the home care plan, there are multiple follow-up questionnaire surveys. Through multiple follow-up surveys of users, data from multiple time periods are collected to form a certain amount of data, so that the system can automatically complete the statistical analysis of hair health-related issues, form effective data to prove the relationship, and display the relevant data to doctors through Echarts charts, thereby helping doctors study the relationship between the diagnosis and treatment of hair health problems and user medication compliance.

[0062] The human hair digital diagnosis and treatment system of the present invention operates in a B / S and C / S coexistence mode. The only difference between managers or doctors and ordinary people is that ordinary people simply use the system and do not need to perform the model training mentioned in the above step S3. The model is trained by the administrator or doctor, and the model used by ordinary users is a mature model that has been trained. The detailed process of the entire system is described as follows.

[0063] The human hair digital diagnosis and treatment system architecture of the present invention is shown in FIG. Figure 3 As shown, including:

[0064] Operating environment layer: This system runs on a server with a Linux operating system installed, and a Docker application container engine is built as the operating environment for the diagnosis and treatment system and its dependencies. Nginx is used as the HTTP and reverse proxy web server, and WSGI is the web server gateway interface. Git is used for version update maintenance and automated deployment of the system code.

[0065] After the detection model is trained, it is deployed in the prepared operating environment and communicated with the application through the message queue to reduce the coupling between the two and improve the performance of the system and the detection model. When the user performs a hair test, the image and electronic questionnaire data of the disease scale are uploaded to the application through the APP. When the application stores the data submitted by the user, it pushes the data to the message queue at the same time. The detection model extracts data for hair detection by detecting changes in the message queue data. The system consultation opinion or intervention opinion finally output by the detection module is transmitted to the application through the message queue. The application stores the data and notifies the user to view the consultation opinion or intervention opinion.

[0066] Data layer: The data layer mainly uses the relational database MySQL as the data storage system, and introduces Redis as a data cache and message queue tool in the system. MySQL, as a data storage system, mainly stores data including basic user information, images and feature data information required for detection, user offline diagnosis information and clinical opinions, system detection result information, family care plan information, and related scheduled task information. The functions of Redis as a data cache and message queue are mainly to accelerate data access, reduce database load, and as a message queue, achieve loose coupling between the application and the detection model, thereby improving the stability and reliability of the system.

[0067] Business logic layer: This detection system uses Django as the program framework for developing back-end business logic functions. At the same time, the detection model is set up independently in an environment, and the Web front-end and business logic functions are relatively separated from the inspection model. The two services communicate and execute by triggering message queues, thereby improving the scalability and reusability of the system.

[0068] View interaction layer: The view interaction layer is the layer that directly contacts the user. The view interaction layer is mainly composed of two parts. The first part is the middle-end management interface based on the B / S architecture, which is mainly implemented by the Reactjs library and the Antd front-end UI component library. At the same time, cornerstone is introduced in the detection model training interface to realize effective annotation operations on the submitted sample images; the second part is the user APP interface based on the C / S architecture, which is mainly implemented using native Android and iOS.

[0069] The functional modules of the human hair digital diagnosis and treatment system of the present invention are shown in FIG. Figure 4 As shown, it includes a user login module, a user management module, a disease scale management module, a hair health detection module, a diagnostic information management module, a file management module, a task management module, a data statistical analysis module, and a follow-up record management module. The use process of the entire system is mainly divided into two types. First, the system administrator or doctor needs to participate in the above-mentioned step S1 to complete the training of the detection model and form a complete training set, so as to obtain a model that can be used by users; second, the user needs to participate in the above steps S2-S5, from submitting hair pictures to participating in follow-up surveys, providing more data to the system, so that the system can statistically analyze and change the effective correlation intensity coefficient from the data, which is convenient for doctors to study hair health.

[0070] Login module: The login module in the Hair Health Detection System mainly provides login for PC users and APP users. Users who log in from the APP are defaulted to ordinary users by the system, and users who log in from the PC are defaulted to administrators and doctors. There are mainly three roles in this system: ordinary users, administrators, and doctors.

[0071] User management module: There are three types of personnel in this system, so personnel with administrator roles are needed to effectively manage the personnel in the entire system to meet the needs of system operation.

[0072] Disease-specific scale management module: Disease-specific scale management mainly involves operations such as querying, adding, updating, deleting, and user submission of questionnaires related to the questionnaire survey involved in the system, and setting the binding coefficient of each questionnaire survey item with the corresponding type in the BASP system classification, so as to facilitate the quantitative processing of the submitted questionnaire data and the next step of execution.

[0073] Hair health detection module: The hair health detection module mainly includes the training of the hair health detection network model and hair health detection. The training branch of the hair health detection network model is only open to doctors and administrators for the training of the detection model. The hair health detection used by ordinary users is a mature detection model trained by doctors or administrators. It only needs to transfer the data submitted by the user to the detection model, and the model will output the diagnosis and intervention opinions to the user after the system detection according to the corresponding data matching.

[0074] Diagnostic information management module: It is used to store the medical information generated after the user obtains the medical data and clinical opinions obtained after offline medical treatment through the file management module, so as to facilitate the user to manage personal diagnostic information. At the same time, the system will also generate a home care plan for the user based on the diagnostic information and generate corresponding scheduled tasks.

[0075] Task management module: The home care plan generated by the system for the user will be recorded in the task module at each step, and the user will be reminded to execute the plan in a timely manner through the mobile phone. The user can also choose to execute the tasks in his or her planned task list according to his or her own situation, and can choose to close the tasks that are not to be executed.

[0076] Follow-up record management module: The system generates a follow-up survey form for each ordinary user at a specified time. Each follow-up survey will be recorded in the system. Users can also obtain a list of follow-up survey record information in which they participated on the APP and view the details.

[0077] Data statistics module: The data statistics module is only developed for administrators and doctors. It mainly counts the hair health test data, diagnosis data, nursing plan status and follow-up survey data submitted by users of this system, obtains a series of data through analysis, and displays the relevant analysis data to doctors, so as to facilitate doctors to study the relationship between hair health, treatment methods and related influencing factors.

[0078] The present invention also provides a human hair digital diagnosis and treatment system for androgenic alopecia, such as Figure 5 As shown, it includes: a data processing module and an information acquisition module electrically connected to the data processing module; the information acquisition module is used to obtain hair images and disease-specific scale questionnaire information; the data processing model outputs medical advice or intervention advice based on the hair images and disease-specific scale questionnaire information.

[0079] The present invention also provides a computer device, comprising: a memory, a processor and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of a human hair digital diagnosis and treatment method for androgenic alopecia.

[0080] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of a human hair digital diagnosis and treatment method for androgenic alopecia.

[0081] Those skilled in the art will appreciate that the embodiments described herein are intended to help readers understand the principles of the present invention, and should be understood that the scope of protection of the present invention is not limited to such specific statements and embodiments. For those skilled in the art, the present invention may have various changes and variations. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of the claims of the present invention.

Claims

1. A human hair detection method for androgenic alopecia, characterized in that: include: A1. Construct a neural network model for hair state detection based on adaptive cropping and multi-granularity feature extraction; A2. Obtain hair images of different states and offline medical records and clinical opinions corresponding to the hair images, and annotate each hair image according to the offline medical records and clinical opinions to obtain a training data set; A3, using the training data set of step A2 to train the hair state detection neural network model constructed in step A1; A4. Input the hair image to be detected into the trained hair state detection neural network model to obtain the hair state detection result.

2. A human hair detection method for androgenic alopecia according to claim 1, characterized in that: The hair state detection neural network model structure described in step S1 includes: The ResNet-50 network is used as the backbone network. The ResNet-50 network includes 4 residual blocks, and an adaptive cropping module is added before the first residual block to achieve automatic cropping of the hair area. It also includes a multi-grain feature network. The adaptive cropping module includes: a 3*3 convolution layer, group normalization, ReLU activation layer and a self-attention module; it is used to output hair area and position mask information; The original image is input into the adaptive cropping module, the image output by the adaptive cropping module is multiplied with the original image and then input into the first residual module, the image output by the first residual block is multiplied with the image output by the adaptive cropping module as the input of the second residual block, the image output by the second residual block is multiplied with the image output by the adaptive cropping module as the input of the third residual block, the image output by the third residual block is multiplied with the image output by the adaptive cropping module as the input of the fourth residual block, so that the four residual blocks respectively output four feature maps of different sizes; The multi-grain feature network includes 4 size transformation modules. The image output by the fourth residual block is used as the input of the first size transformation module; the image output by the third residual block and the image output by the first size transformation module are used as the input of the second size transformation module; the image output by the second residual block and the image output by the second size transformation module are used as the input of the third size transformation module; the image output by the first residual block and the image output by the third size transformation module are used as the input of the fourth size transformation module; The images output by the four size transformation modules are input into a fully connected layer, and the fully connected layer outputs the final hair detection results.

3. A human hair detection method for androgenic alopecia according to claim 2, characterized in that: The hair detection results include at least: average color, texture, shape, edge length and outline, and glossiness.

4. A digital diagnosis and treatment method for human hair for androgenic alopecia, characterized in that: include: B1. Data acquisition steps include: B11. Using the human hair detection method for androgenic alopecia according to claim 1, obtaining the hair detection result of the hair image to be detected; B12. Obtain the data of the disease-specific questionnaire, which includes: hair density, hair diameter, hair follicle health, genetic history, eating habits, living habits, mental stress level and medication status; B2. Data preprocessing steps include: B21. Use Cronbach's coefficient to quantify the data of the disease-specific scale questionnaire to obtain quantitative data; B22, classifying the hair detection result of step B11 and the quantitative data obtained in step B21 according to BASP, marking them as feature selection criteria, and obtaining a feature data set; B23. Calculate the information entropy of each feature in the feature data set, and calculate the information gain based on the information entropy; B3. Data analysis step: information gain is used to provide a basis for decision tree generation for the C4.5 decision tree algorithm, thereby generating a diagnosis and treatment opinion and intervention opinion decision tree with different intervention levels for different types and severity of hair conditions.

5. A human hair digital diagnosis and treatment method for androgenic alopecia according to claim 4, characterized in that: During the decision tree generation process, the feature with the largest information gain in each category is selected as the partition feature.

6. A digital diagnosis and treatment method for human hair for androgenic alopecia according to claim 4, characterized in that: The content of the disease-specific scale questionnaire in step S5 includes: hair density, hair diameter, hair follicle health, genetic history, eating habits, living habits, mental stress level and medication status.

7. A digital diagnosis and treatment method for human hair for androgenic alopecia according to claim 4, characterized in that: The Cronbach's coefficient calculation formula is: Among them, α is the reliability coefficient, K is the number of test questions, represents the score variation of all subjects on question i, is the variance of the total scores of all subjects.

8. A human hair digital diagnosis and treatment system for androgenic alopecia, comprising: A data processing module and an information acquisition module electrically connected to the data processing module; the information acquisition module is used to acquire hair images and disease-specific scale questionnaire information; The data processing module outputs a consultation opinion or an intervention opinion based on the diagnosis and treatment method described in claim 4 according to the hair image and the disease-specific scale questionnaire information.

9. A computer device, characterized in that: include: A memory, a processor and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the diagnosis and treatment method described in claim 4.

10. A computer-readable storage medium, characterized in that: A computer program is stored thereon, and when the computer program is executed by a processor, the steps of the diagnosis and treatment method described in claim 4 are implemented.

Citation Information

Patent Citations

  • Questionnaire survey method, questionnaire survey system and electronic device

    CN108920677A

  • AI-assisted male alopecia areata Hamiltonian classification prediction analysis method for AGA clinical image

    CN110033448A

  • Hair follicle detection method based on deep convolutional neural network

    CN112308827A

  • Intelligent alopecia decision-making method and device based on AI and computer readable storage medium

    CN112687392A

  • Scalp type diagnosis system based on scalp state information and scalp improvement method using same

    CN117177707A

Cited By

  • Androgenic alopecia detection method and system based on artificial intelligence

    CN121190403A