Clinical decision support system for diabetic foot wound treatment

Abstract

The invention provides a diabetic foot wound treatment clinical decision support system, which comprises a data input module used for inputting patient information during first recording; the image processing module is used for acquiring and processing a plurality of images of the foot wound; the wound grading module is used for grading wounds according to the image processing result of the image processing module; and the clinical decision-making module is used for making a next clinical decision-making scheme by utilizing machine learning or deep learning according to the examination result of the patient and the wound classification. The diabetic foot wound treatment clinical decision support system can realize automatic measurement, automatic treatment and automatic analysis of foot ulcers, avoids limitation of manual treatment in the prior art, and can provide favorable decision support for treatment of diabetic foot ulcers.

Description

Technical Field

[0001] The present invention relates to the field of medical health technology, and in particular to a clinical decision support system for treating diabetic foot wounds. Background Art

[0002] Diabetic foot ulcers (DFUs) are serious complications of diabetes. They are primarily caused by chronic high blood sugar levels, which damage nerves and blood vessels, leading to decreased sensation in the feet, muscle atrophy, and hyperkeratosis, making them more susceptible to ulceration and infection. DFUs not only prolong hospital stays and reduce the quality of life of diabetic patients, but also incur significant medical costs.

[0003] Wound management for diabetic foot ulcers plays a vital role in preventing infection, promoting healing, reducing amputation rates and mortality, and improving patients' quality of life. In addition to normal wound management, timely recording of treatment results and progress is also necessary when managing diabetic foot ulcers.

[0004] Diabetic foot ulcers are chronic wounds. Accurate wound measurement and assessment are crucial for healthcare professionals to better understand their evolution, which is crucial for determining care and treatment plans. Wound measurement and assessment primarily involves monitoring wound area, depth, volume, color, and other aspects. Wound indicators directly reflect the degree of wound healing and, combined with other patient health data, provide more effective clinical decision support. Furthermore, care and treatment plans can be adjusted based on the patient's specific needs.

[0005] Currently, wound measurement mainly relies on physical measurement methods, such as cotton swabs, probes, or rulers. This measurement method directly contacts the patient's wound, and the measurement accuracy is unstable, which increases the patient's pain and the risk of infection. Wound treatment requires manual participation and manual recording throughout the process. For example, every time the wound is changed, after cleaning the wound, the patient's bed number, name, and medical ID number are marked with a wound ruler, and then the wound ruler is placed on the patient's wound. The staff member takes a photo with their own mobile phone and uploads it to the diabetes WeChat work group, so that the doctor can check the patient's wound progress or recovery in time. This manual recording method will lead to: (1) the picture expires and is not saved; (2) it occupies the staff member's mobile phone memory; (3) the doctor only sees the size of the wound and the area of ​​infection, and cannot obtain the wound dressing plan and cannot evaluate the effectiveness of the dressing plan; (4) it cannot intelligently match the uniformity of the dressing plan for wounds with similar diabetic foot wound grades and infection levels. Summary of the Invention

[0006] In order to solve the above problems, the present invention provides a clinical decision support system for diabetic foot wound treatment, which can realize automatic measurement, automatic treatment and automatic analysis of diabetic foot ulcers, avoid the limitations of manual treatment, and provide favorable decision support for the subsequent treatment of diabetic foot ulcers.

[0007] The present invention provides a clinical decision support system for treating diabetic foot wounds, comprising: The data entry module is used to enter the patient's name, age, ID number, and medical ID number when first recording. After entering the identity information, the system will automatically associate and import the patient's medical history and previous examination results. It provides an API interface compatible with the hospital HIS system to achieve multi-source data integration. In subsequent processing, login and query can be performed using any identity information; Image processing module, high-definition camera with ruler to obtain multi-angle images of foot wounds (see Figure 2 ), with a scale accuracy of ≤0.1mm, to obtain multiple images of the foot wound; based on the Mask R-CNN algorithm, the wound edge is segmented, and the area (error rate <3%), depth (assisted by laser ranging), and volume (3D reconstruction technology) are calculated. The HSV color model is used to identify red / yellow / black / mixed color areas and quantify the proportion of necrotic tissue; finally, a comprehensive score for the foot wound is obtained based on the preset weight ratio; The wound grading module grades wounds according to the image processing results of the image processing module. Different scores correspond to different wound grades. The wound grades include: Grade 0: obvious blood supply deficiency in the foot, but no open wound; Grade 1: superficial ulcers on the foot; Grade 2: ulcers even with tendon or joint capsule exposure; Grade 3: deep ulcers accompanied by osteomyelitis or abscess formation; Grade 4: wet or dry gangrene may have cellulitis; Grade 5: extensive gangrene occurs, high paraplegia; the total score is calculated by weighting the area (30%), depth (25%), color (20%), nerve damage (15%), and vascular status (10%) to establish a comprehensive scoring model.

[0008] The clinical decision-making module is used to make the next clinical decision plan based on patient examination results and wound classification using machine learning or deep learning.

[0009] Preferably, the next clinical decision-making process using machine learning or deep learning includes: Preprocess patient examination results and wound grading data, remove missing values ​​and outliers, and perform data normalization; Divide the dataset into a training set and a test set, select a large AI model, set appropriate model parameters, and then train the model. For the trained model, use cross-validation to evaluate the model's stability and accuracy. Adjust model parameters and repeat the training and validation steps until satisfactory training results are achieved; The trained model is used to provide further decision-making solutions.

[0010] Preferably, a plurality of foot wound images with scales are acquired, and the image processing software obtains the area of ​​the foot wound by pixel extraction, and calculates the volume of the foot wound by the area and depth of the foot wound.

[0011] Preferably, the foot wound colors identified by the image processing software include red, yellow, black and mixed colors.

[0012] Preferably, the same weight is set for the area, depth, color, nerve damage, and blood vessel damage of the foot wound.

[0013] Preferably, the system further comprises: A data storage module for storing the patient's identity information, medical history, examination results, wound images, wound grade, and historical treatment plans; A data retrieval module is used to retrieve the patient's identity information, medical history, examination results, wound images, wound grade, and historical treatment plans; The statistical analysis module is used to perform statistical analysis on the stored data.

[0014] The clinical decision support system for treating diabetic foot wounds provided by the present invention can realize automatic measurement, automatic treatment and automatic analysis of foot ulcers, avoiding the limitations of manual treatment in the existing technology and providing favorable decision support for the treatment of diabetic foot ulcers. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 This is a framework diagram of a clinical decision support system for diabetic foot wounds according to the present invention; Figure 2 is a schematic diagram of an image of a foot wound with a ruler. DETAILED DESCRIPTION The embodiments of the present invention are described below with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are only illustrative and not restrictive.

[0016] like Figure 1 As shown, the present invention provides a clinical decision support system for diabetic foot wound treatment, which includes a data entry module, an image processing module, a wound grading module, a clinical decision module, a data storage module, a data retrieval module and a statistical analysis module.

[0017] The data entry module is used to enter the patient's name, age, ID number, and medical ID number when recording for the first time. After entering the identity information, the system will automatically associate and import the patient's medical history and previous examination results. In subsequent processing, login and query can be performed through any identity information.

[0018] The image processing module is used to obtain multiple images of foot wounds, analyze them using image processing software, calculate the area, depth, color, nerve damage and vascular damage of the foot wounds, and obtain a comprehensive score of the foot wounds based on a preset weight ratio; the obtained area, depth, color, nerve damage and vascular damage of the foot wounds and the comprehensive score of the foot wounds can be used as image processing results for image analysis by the image processing module.

[0019] The wound grading module is used to grade wounds based on the image processing results of the image processing module. Different scores correspond to different wound grades. The wound grades include: Grade 0: obvious blood supply deficiency in the foot, but no open wound; Grade 1: superficial ulcers on the foot; Grade 2: ulcers even with tendon or joint capsule exposure; Grade 3: deep ulcers accompanied by osteomyelitis or abscess formation; Grade 4: wet or dry gangrene may have cellulitis; Grade 5: extensive gangrene occurs, with high paraplegia.

[0020] The clinical decision-making module is used to make the next clinical decision plan based on patient examination results and wound classification using machine learning and deep learning.

[0021] The clinical decision support system for treating diabetic foot wounds of the present invention can realize automatic measurement, automatic treatment and automatic analysis of foot ulcers, avoiding the limitations of manual treatment in the prior art and providing favorable decision support for the treatment of diabetic foot ulcers.

[0022] In some optional embodiments, the clinical decision module utilizes machine learning and deep learning to formulate the next clinical decision plan, including: First, the patient's examination results and wound grading data are preprocessed to remove missing values ​​and outliers and perform data normalization. Examination results include the corresponding results of various patient-related medical examinations imported into the data entry module; wound grading data includes image processing results and corresponding wound grading. A dataset can be generated for these preprocessed examination results and wound grading data for model training. For example, examination result data preprocessing can include: data cleaning, such as removing irrelevant symbols: deleting special characters, punctuation marks, and other irrelevant information from the examination results to reduce the impact of noise data on subsequent analysis; error correction: correcting typos and spelling errors in text data to ensure data accuracy. Data standardization: Standardizing the numerical data in the examination results to uniformly measure and scale, facilitating subsequent analysis and modeling.

[0023] The preprocessing of wound grading data can include: image normalization, standardizing the wound images; data enhancement, applying data enhancement techniques such as random cropping, rotation, color transformation, horizontal and vertical flipping, etc., to enrich the diversity of training samples and improve the robustness and generalization ability of the model; segmenting the wound images to extract the wound area for more accurate wound grading, etc.

[0024] Next, the dataset is divided into a training set and a test set. An AI large model is selected and appropriate model parameters are set for model training. Cross-validation is used to evaluate the stability and accuracy of the trained model. The AI ​​large model is constructed based on traditional machine learning models such as logistic regression, decision trees, random forests, and support vector machines, or based on deep learning models such as convolutional neural networks, recurrent neural networks, and deep learning networks, although this embodiment does not limit this.

[0025] Again, adjust the model parameters and repeat the training and validation steps until satisfactory training results are achieved; Finally, the trained model is used to provide further decision-making solutions.

[0026] By constructing and training a large AI model, the embodiment of the present invention can realize the automated processing of wound judgment, accurately grasp the patient's wound condition at different stages, greatly improve the efficiency of wound status judgment, and provide effective auxiliary work for subsequent work.

[0027] As previously mentioned, the image processing module can perform image analysis on multiple acquired images. Furthermore, multiple foot wound images with a ruler are acquired. The image processing software extracts the area of ​​the foot wound through pixel extraction and calculates the volume of the foot wound based on the area and depth of the foot wound. For example, drag a wound image into the software; adjust the image angle to ensure that the image meets the required angle, with the ruler horizontal as the standard; adjust the distance between the ruler and the wound to ensure that the ruler is 2-5 cm long, the upper edge of the ruler is approximately 0.5 cm from the wound, the wound is centered, and the image is approximately square; adjust the image brightness and contrast to clearly display the wound area; adjust the color threshold to select the wound area by changing the threshold to cover the entire wound area as accurately as possible; use the Magic Wand tool to click on the selected area to automatically generate a region of interest and save it in the ROI manager; set a standard distance: select the Line tool from the menu, draw a distance on a ruler of known length, and then set the standard distance; calculate the area: select the image and the corresponding ROI in the ROI manager, click Measure to measure, and save the measurement data as a .csv file.

[0028] Convolutional neural networks (CNNs) can be used to extract and identify features from wound images. They can be trained to identify wound edges and features, thereby estimating wound depth. For example, the Mask R-CNN model can be used for automatic wound segmentation and detection, a method that has demonstrated excellent performance in clinical wound image analysis, achieving high accuracy.

[0029] In addition, a laser ranging sensor can also be used to obtain the wound depth.

[0030] Furthermore, the image processing software identifies foot wound colors, including red, yellow, black, and mixed colors. Red indicates healthy granulation tissue within the wound, commonly seen in clean or healing wounds (in the healing process); yellow indicates slough, exudate, or infection within the wound (the wound is not yet ready to heal); black indicates necrotic tissue lacking blood supply, soft or hard scabs (the wound has no tendency to heal); mixed colors indicate tissues of different colors, and the proportion of each color can be described as a percentage.

[0031] The same weight can be set for the area, depth, color, nerve damage, and vascular damage of the foot wound. For example, the area, depth, color, nerve damage, and vascular damage of the foot wound each account for 20 points, and the final grade of the foot wound is determined based on their sum.

[0032] The following is an example of wound depth measurement provided by an embodiment of the present invention.

[0033] 1. Data Collection and Equipment Configuration Hardware: A Microsoft Azure Kinect DK TOF camera (resolution 1024 × 1024, depth accuracy ±0.5 mm) was used, mounted on an adjustable tripod, 20–30 cm from the wound surface at a vertical angle of ±5°. Environmental requirements: Shadowless lighting (color temperature 5500K), avoid reflection interference; During the shooting, the patient lies in a supine position, the wound area is exposed, and a transparent sterile film is covered after disinfection; the wound is taken as the center and 8 angles are shot around it (one frame every 45°) to ensure that all edges are covered; the ambient temperature and humidity (25±2℃, humidity 50%-60%) are recorded simultaneously to calibrate the data.

[0034] 2. 3D Reconstruction and Registration Preprocessing: Use the Open3D library to denoise the original point cloud (statistical outlier removal, number of neighborhood points = 50, standard deviation multiple = 1.5); Scale calibration: Correct camera distortion using a calibration plate (chessboard accuracy 0.1mm); ICP registration: The iterative closest point (ICP) algorithm is used, with the initial view as the reference, the iterative error threshold set to 0.1 mm, and the maximum number of iterations is 100. After registration, a fused point cloud is generated with a resolution of 0.2 mm / pixel.

[0035] 3. Deep Calculation and Verification Definition of reference plane: A 5mm healthy skin area around the wound was selected and the plane equation ax+by+cz+d=0 was fitted using the RANSAC algorithm; Calculation of the concave value: Project the wound point cloud along the Z axis to the reference plane, and calculate the maximum vertical distance Dmax = max(z reference plane - z wound); Output depth report: including mean, maximum and 3D heat map (see Figure 2 ); Accuracy verification: Comparison with the gold standard: 50 patients were selected for comparison with CT scan results (slice thickness 0.625 mm). Bland-Altman analysis showed that the 95% consistency limit was ±0.48 mm. Clinical case: Patient B (male, 62 years old, grade 2 ulcer), TOF measurement depth was 4.2mm, CT result was 4.0mm, with an error of 4.8%.

[0036] In an optional embodiment of the present invention, the clinical decision support system for diabetic foot wound treatment may further include: The data storage module is used to store the patient's identity information, medical history, examination results, wound images, wound grade and historical treatment plans.

[0037] The data storage module allows for long-term data storage, eliminating the issue of images being outdated and inaccessible due to recording them via chat software. Treatment personnel can share and search for treatment plans, gaining timely understanding of wounds and wound dressing plans, allowing them to adjust treatment plans promptly.

[0038] The database stored in the data storage module is editable, allowing for intelligent categorization and chronological display. Wound care professionals can annotate and describe each patient's treatment and dressing change plan. The database also features a search function, allowing them to easily access wound data from past and current patients.

[0039] The data retrieval module is used to retrieve the patient's identity information, medical history, examination results, wound images, wound grade and historical treatment plans. The clinical decision support system for the treatment of diabetic foot wounds in this embodiment can provide a unified interface to the outside world. In actual applications, when providing retrieval, you can search by keywords, such as by obtaining the search text and further performing keyword matching, and the system searches the database for examination results or wound grade records containing these keywords. For example, if the user enters "diabetic foot wound", the system will retrieve all relevant data containing "diabetic foot wound" in the description. In addition to exact matching, fuzzy matching technology can also be used to improve the flexibility of retrieval. For example, if the user enters "diabetic wound", the system can not only retrieve "diabetic foot wound", but also retrieve similar data such as "diabetic hand wound". It can also be based on classification retrieval, full-text retrieval or intelligent recommendation retrieval, which is not limited in this embodiment.

[0040] The statistical analysis module is used to perform statistical analysis on stored data. It can be connected to the data storage module. Its primary function is to organize, analyze, and interpret collected data to help users better understand the information contained in the data and support decision-making. It typically includes descriptive statistics, inferential statistics, and visualization of data.

[0041] The data transmission module is used to upload data to the cloud or mobile phone. Specifically, data transmission can be achieved by wired or wireless means.

[0042] The diabetic wound treatment clinical decision support system of the embodiment of the present invention also has the following functions: Automatic matching function: When a new patient changes their dressing, after taking a wound photo, data retrieval can be automatically performed to match corresponding or highly similar case data records, and a quick decision on the treatment plan can be made; The function of exporting patient wound treatment pictures can be used to make PPTs for easy sharing and communication; Data cloud upload function: after taking wound pictures, they can be automatically uploaded to the system cloud or mobile phone to avoid picture deletion or unsaved; Data statistics function: statistics on the number of people who changed dressings annually, the number of times dressings were changed annually, the number of people who were cured, the number of times they were cured, etc. It can also analyze the healing speed of the same type of wounds.

[0043] The process of generating a processing solution is described below through an embodiment.

[0044] 1. Data preparation and model building Dataset source: Multicenter collaborative data (Hospital A, Hospital B, etc.), a total of 10,372 DFU patients, spanning 2018-2023; Data fields: wound parameters (area, depth, color ratio), biochemical indicators (HbA1c, CRP), treatment plan (dressing type, debridement method), healing time (days); Data cleaning: Missing value processing: KNN interpolation (k = 5) was used for continuous variables, and fields with missing values ​​> 30% were deleted for categorical variables; outlier elimination: Based on the Tukey method (IQR = 1.5), extreme values ​​with wound area > 100 cm² or depth > 30 mm were deleted; Feature engineering: Standardization: Z-score normalization (mean = 0, standard deviation = 1); Time series features: For patients returning for follow-up visits, extract the parameter change rate of the first three dressing changes.

[0045] 2. Model training and optimization Algorithm selection: XGBoost: for static feature prediction (learning rate = 0.01, tree depth = 6, boosting rounds = 500); LSTM: for time series data analysis (hidden layers = 128, time steps = 3, dropout = 0.2); Training process: First, divide the dataset into 70% training set, 15% validation set, and 15% test set; second, cross-validation: 5-fold cross-validation, optimize hyperparameters (such as regularization coefficient λ = 0.1); third, ensemble strategy: weighted voting (XGBoost weight 60%, LSTM weight 40%); Performance evaluation: AUC calculation: based on the ROC curve, with a threshold from 0 to 1 in steps of 0.01; specificity = true negative / (true negative + false positive), sensitivity = true positive / (true positive + false negative); test set results: AUC = 0.92 (95% CI: 0.89-0.94), specificity 85.3%, sensitivity 89.1%.

[0046] 3. Solution output and case studies Rule engine: Based on the IWGDF guidelines, set the priority: infection control > blood supply improvement > wound repair; example logic: If CRP > 10 mg / L and wound score ≥ grade 3, intravenous antibiotics are mandatory; dynamic recommendation: Patient C (female, 58 years old, grade 3 ulcer, HbA1c = 9.0%): Model input: area = 8.6 cm², depth = 6.5 mm, yellow necrosis 45%, CRP = 15 mg / L; Output scheme: (1) Emergency treatment: broad-spectrum antibiotics (cefotaxime 2 g / d for 7 consecutive days); (2) Local treatment: ultrasonic debridement (twice a week) + collagen dressing; (3) Metabolic regulation: Intensive insulin treatment (target fasting blood glucose <7 mmol / L); Follow-up results: After 4 weeks, the area was reduced to 3.2 cm² and CRP dropped to 5 mg / L.

[0047] The implementation effects and verification are as follows: Efficiency comparison: Traditional manual recording takes 30 minutes per case, while the entire process of this system takes ≤ 5 minutes; Cost analysis: The average annual number of dressing changes per patient was reduced from 18 to 12, saving approximately ¥2,300 in consumables. Clinical adoption rate: The adoption rate in pilot hospitals reached 92% within 3 months, and was well received by medical staff for its "intuitive operation" and "reliable decision-making."

[0048] The clinical decision support system for diabetic wound treatment in an embodiment of the present invention is equipped with multiple functional modules and has functions such as data storage, retrieval, analysis, data transmission, data statistics, and visual display. It can not only store patient treatment data in a timely manner, but also save staff mobile phone memory, and accurately understand the patient's wound dressing plan, thereby effectively evaluating the treatment accuracy and realizing the uniformity of intelligent matching of diabetic foot wound grading and wound dressing plans with similar infection degree grading.

[0049] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present invention should be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

Claims

1. A clinical decision support system for diabetic foot wound management, characterized by: include: The data entry module is used to enter the patient's name, age, ID number, and medical ID number when recording for the first time. After entering the identity information, the system will automatically associate and import the patient's medical history and previous examination results. It provides an API interface compatible with the hospital HIS system to achieve multi-source data integration; An image processing module is used to obtain multiple images of the foot wound, analyze and calculate the area, depth, color, nerve damage and blood vessel damage of the foot wound, and obtain a comprehensive score for the foot wound based on a preset weight ratio; The wound grading module is used to grade wounds based on the image processing results of the image processing module. Different scores correspond to different wound grades. The wound grades include: Grade 0: obvious blood supply deficiency in the foot, but no open wound; Grade 1: superficial ulcers on the foot; Grade 2: ulcers even with tendon or joint capsule exposure; Grade 3: deep ulcers with osteomyelitis or abscess formation; Grade 4: wet or dry gangrene with possible cellulitis; Grade 5: extensive gangrene and high paraplegia. The clinical decision-making module is used to make the next clinical decision plan based on patient examination results and wound classification using machine learning or deep learning.

2. The clinical decision support system for diabetic foot wound management according to claim 1, characterized in that: The clinical decision module uses machine learning or deep learning to formulate the next clinical decision plan, including: Preprocess patient examination results and wound grading data, remove missing values ​​and outliers, and perform data normalization; Divide the dataset into a training set and a test set, select a large AI model, set appropriate model parameters, and then train the model. For the trained model, use cross-validation to evaluate the model's stability and accuracy. Adjust model parameters and repeat the training and validation steps until satisfactory training results are achieved; The trained model is used to provide further decision-making solutions.

3. The clinical decision support system for diabetic foot wound management according to claim 1, characterized in that: A plurality of foot wound images with rulers are obtained, and the image processing software obtains the area of ​​the foot wound by pixel extraction, and calculates the volume of the foot wound by the area and depth of the foot wound.

4. The clinical decision support system for diabetic foot wound management according to claim 1, characterized in that: The foot wound colors identified by the image processing software include red, yellow, black and mixed colors.

5. The clinical decision support system for diabetic foot wound management according to claim 1, characterized in that: Set equal weights for foot wound area, depth, color, nerve damage, and vascular damage.

6. The clinical decision support system for diabetic foot wound management according to claim 1, characterized in that: The system further comprises: A data storage module for storing the patient's identity information, medical history, examination results, wound images, wound grade, and historical treatment plans; A data retrieval module is used to retrieve the patient's identity information, medical history, examination results, wound images, wound grade, and historical treatment plans; Statistical analysis module, used to perform statistical analysis on stored data; Data transmission module, used to upload data to the cloud or mobile phone.

Images