Knee osteoarthritis treatment decision tree system and device

Abstract

The invention provides a knee osteoarthritis treatment decision-making tree system and device, and relates to the technical field of knee osteoarthritis treatment decision-making trees, and the system comprises a data receiving module, a data processing module and a result generation module. The data receiving module is used for receiving natural language text data and iconography image data of the patient and respectively preprocessing the natural language text data and the iconography image data; the data processing module comprises a text feature unit, an image feature unit and a decision unit; the decision-making unit is used for fusing the text features and the image quantization features to obtain comprehensive features; based on the comprehensive features, through a decision tree model, a treatment scheme for knee osteoarthritis is output; and the result generation module is used for sending out the treatment scheme. According to the scheme, information such as medical records of patients can be automatically analyzed, personalized treatment schemes are given, interference of human factors is reduced, real-time and objective decision support is provided for doctors, and therefore the diagnosis and treatment efficiency and effect are improved.

Description

Technical Field

[0001] This invention relates to the field of decision tree technology for the treatment of knee osteoarthritis, and in particular to a decision tree system and device for the treatment of knee osteoarthritis. Background Technology

[0002] Knee osteoarthritis, a common joint disease, significantly impacts patient prognosis due to the accuracy of its diagnosis and treatment decisions. Traditional treatment models rely heavily on individual physician experience, leading to low diagnostic efficiency and susceptibility to subjective factors when dealing with complex and diverse patient information. With the explosive growth of medical data, effectively integrating and utilizing patient medical records and imaging data to provide physicians with precise, scientific, and objective treatment decision support has become a critical issue that urgently needs to be addressed. Summary of the Invention

[0003] The purpose of this invention is to provide a decision tree system and device for the treatment of knee osteoarthritis, so as to solve at least one of the above-mentioned technical problems existing in the prior art.

[0004] In a first aspect, to solve the above-mentioned technical problems, the present invention provides a decision tree system for the treatment of knee osteoarthritis, including a data receiving module, a data processing module and a result generation module; The data receiving module is used to receive the patient's natural language text data and imaging image data, and to preprocess both of them respectively; The data processing module includes a text feature unit, an image feature unit, and a decision unit; The text feature unit extracts text features related to knee osteoarthritis from natural language text data using natural language processing (NLP) methods; it then performs structured processing on the text features; thus, it can effectively transform natural language text data into structured feature data and establish a connection with disease diagnosis and treatment decisions, providing accurate input data for subsequent decision tree models; The image feature unit identifies and segments regions of interest (such as the tibia, femur, anterior cruciate ligament, and meniscus) in radiographic image data using a convolutional neural network to obtain segmentation results. Based on the segmentation results, it extracts morphological features (such as morphological angles), texture features (such as joint degeneration and bone spurs) and intensity features related to knee osteoarthritis and performs quantitative analysis to obtain image quantitative features. The decision unit fuses text features and image quantization features to obtain comprehensive features; based on the comprehensive features, it outputs a treatment plan for knee osteoarthritis through a decision tree model. The result generation module is used to outsource the treatment plan; In this way, by using artificial intelligence methods such as machine learning to automatically analyze the medical data of patients with knee osteoarthritis, personalized treatment plans can be obtained quickly and accurately, avoiding the influence of doctors' subjective factors and thus improving the prognosis.

[0005] In one feasible implementation, the natural language text data includes electronic medical records, doctor's records, and medical history questionnaires; the electronic medical record includes the patient's basic information, symptom description, physical examination, imaging examination reports, and treatment plan; the basic information includes age and gender; the symptom description includes knee pain level and knee joint deformity status; the physical examination includes body mass index (BMI) testing, knee tenderness testing, knee joint range of motion testing, knee joint stability testing, knee joint deformity testing, thigh circumference testing, and quadriceps muscle strength testing; the BMI testing is an indicator of whether weight is healthy relative to height; the knee tenderness testing refers to the pain induced by applying pressure to specific anatomical locations of the knee joint and is evaluated according to pain level; the knee joint range of motion testing refers to the ability of the knee joint to move actively or passively. The range of flexion and extension angles is assessed using a goniometer; the knee joint stability test refers to the knee joint's ability to maintain its normal anatomical position and function under load or stress, mainly assessed through the anterior drawer test, posterior drawer test, Lachman test, varus / valgus stress test, and axis shift test, and evaluated according to stability grading; the knee joint deformity test refers to the degree of deviation of the knee joint from the normal anatomical axis in the coronal or sagittal plane, measured visually or with a ruler; the thigh circumference test refers to the diameter measurement at a fixed distance above the upper edge of the patella using a soft ruler; the quadriceps muscle strength test assesses the contractile strength of the quadriceps muscle, including isometric muscle strength in the horizontal and vertical directions of the leg, and the testing methods include manual muscle strength testing, functional testing, and instrumental quantitative testing.

[0006] In one feasible implementation, the radiological image data includes X-ray films, CT images, and MRI images, etc.

[0007] In one feasible implementation, a method for preprocessing natural language text data includes: Using Python tools, regular expressions were written using the NLTK and Spacy libraries to remove garbled characters, special symbols, and other interfering information from electronic medical records, thereby ensuring the cleanliness of natural language text data and improving the accuracy of subsequent text processing. The Pandas library was used to structure the natural language text data. By extracting key information, including patient ID, consultation date, and text type, unstructured data in the natural language text data was transformed into structured data, facilitating subsequent analysis and processing. The natural language text data was then categorized and stored according to dimensions such as patient ID, consultation date, and text type before being imported into the database, ensuring data traceability and efficiency.

[0008] In one feasible implementation, a specific method for preprocessing radiographic image data includes: Based on Python tools, histogram equalization, median filtering, and affine transformation are performed on X-ray images using OpenCV library functions; irrelevant information is removed from CT images and image quality is improved by thresholding; and noise reduction and intensity normalization are performed on MRI images. Reconstructing the volume of a 2D slice in 3D.

[0009] In one feasible implementation, the natural language processing method includes: By using the pre-trained GloVe word vector model, words in natural language text data are transformed into word vectors with semantic and contextual relevance. This not only preserves the semantic information of words, but also captures the co-occurrence relationship between words, which is convenient for subsequent model training and inference. By combining medical professional dictionaries and using the BIO (Begin, Inside, Outside) annotation mode, medical terms in natural language text data are annotated and entity recognized, thereby effectively marking the boundaries of medical terms. This enables the system to identify medical entities such as "knee pain", "joint locking", and "joint swelling" and mark their positions in the text. For the field of knee joint diseases, the BERT model was fine-tuned to adapt to the specific context and medical terminology of the knee joint disease field; the BERT model was used to extract text features related to knee osteoarthritis from natural language text data.

[0010] In one feasible implementation, fine-tuning the training includes: dividing the natural language text data into a training set, a validation set, and a test set according to a preset ratio; selecting a cross-validation method; using the cross-entropy loss function as the optimization objective; adjusting the model parameters through the backpropagation algorithm to minimize the prediction error rate; and setting hyperparameters such as the learning rate and batch size to ensure training stability and convergence speed.

[0011] In one feasible implementation, a deep learning model can be used to replace the BERT model to extract more specific text features, such as "pain location", "pain nature", "pain intensity" and "pain duration", so that the system can more comprehensively and thoroughly assess the patient's symptom presentation.

[0012] In one feasible implementation, the convolutional neural network is a U-Net model architecture, specifically including an encoder (downsampling path) and a decoder (upsampling path). The encoder extracts high-level semantic features of the image step by step through convolutional layers and max pooling layers; The decoder improves segmentation accuracy by upsampling high-level semantic features corresponding to the encoder's level.

[0013] In one feasible implementation, the training process of the convolutional neural network includes: using the ITK-SNAP tool to annotate the regions of interest in the radiographic images as the gold standard; dividing the annotated radiographic image data into training and validation sets according to a preset ratio; using Categorical Cross-Entropy Loss, Dice Loss, and Boundary Loss as loss functions; selecting the Adam optimizer; and training the convolutional neural network after setting the network parameters.

[0014] In one feasible implementation, the decision tree model is specifically trained using a decision tree algorithm; feature splitting points are selected based on cost functions such as information gain and Gini index to progressively divide the input features; the decision tree model is evaluated and optimized using performance indicators such as accuracy, recall, and F1 score; and the decision-making process is displayed through visualization methods, thereby providing doctors with intuitive decision-making references.

[0015] In one feasible implementation, the treatment regimen includes drug therapy and / or physical therapy and / or surgical intervention.

[0016] Secondly, based on the same inventive concept, this application also provides a knee osteoarthritis treatment decision tree device, including a processor, a memory, and a bus. The memory stores instructions and data read by the processor, and the processor is used to call the instructions and data in the memory to realize the knee osteoarthritis treatment decision tree system as described above. The bus connects the various functional components for transmitting information.

[0017] By adopting the above technical solution, the present invention has the following beneficial effects: The present invention provides a decision tree system and device for the treatment of knee osteoarthritis, which can automatically analyze patients' medical records and other information, provide personalized treatment plans, reduce the interference of human factors, and provide doctors with real-time, scientific and objective decision support, thereby improving the efficiency and effectiveness of diagnosis and treatment. Attached Figure Description

[0018] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0019] Figure 1 A decision tree system diagram for the treatment of knee osteoarthritis provided in this embodiment of the invention; Figure 2 This is an architecture diagram of a skip connection provided in an embodiment of the present invention; Figure 3 This is an illustration of the hip-knee-ankle angle provided for an embodiment of the present invention; Figure 4 Explanatory diagrams of the lateral angle of the distal femur and the medial angle of the proximal tibia provided in embodiments of the present invention; Figure 5 Explanation diagram of femoral-tibia angle provided for embodiments of the present invention; Figure 6 This is a schematic diagram showing the location of the main cartilages in the knee joint, provided in an embodiment of the present invention. Figure 7 This is a schematic diagram showing the location of the main ligaments of the knee joint provided in an embodiment of the present invention; Figure 8 This is an example diagram of a decision tree model provided in an embodiment of the present invention. Detailed Implementation

[0020] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] In the description of this invention, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0022] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0023] The present invention will be further explained below with reference to specific embodiments.

[0024] It should also be noted that the specific embodiments or implementation methods described below are a series of optimized settings listed by the present invention to further explain the specific content of the invention, and these settings can be combined or used in conjunction with each other.

[0025] Example 1: like Figure 1 As shown in the figure, this embodiment provides a decision tree system for the treatment of knee osteoarthritis, including a data receiving module, a data processing module, and a result generation module; The data receiving module is used to receive natural language text data and imaging image data of patients with (knee osteoarthritis), and to preprocess both of them respectively. The data processing module includes a text feature unit, an image feature unit, and a decision unit; The text feature unit extracts text features related to knee osteoarthritis from natural language text data (such as symptom descriptions, medical history, physical examinations, medication use, and previous treatments) using natural language processing (NLP) methods. These text features are then structured using methods such as word segmentation, entity recognition, and relation extraction. This effectively transforms natural language text data into structured feature data, establishing a connection with disease diagnosis and treatment decisions, and providing accurate input data for subsequent decision tree models. The image feature unit identifies and segments the region of interest in the radiographic image data through a convolutional neural network to obtain the segmentation result; based on the segmentation result, it extracts morphological features, texture features and intensity features related to knee osteoarthritis and then performs quantitative analysis to obtain image quantitative features. The decision unit fuses text features and image quantization features to obtain comprehensive features; based on the comprehensive features, it outputs a treatment plan for knee osteoarthritis through a decision tree model. The result generation module is used to outsource the treatment plan; In this way, by using artificial intelligence methods such as machine learning to automatically analyze the medical data of patients with knee osteoarthritis, personalized treatment plans can be obtained quickly and accurately, avoiding the influence of doctors' subjective factors and thus improving the prognosis.

[0026] Furthermore, the natural language text data includes electronic medical records, doctor's records, and medical history questionnaires; the electronic medical records include the patient's basic information, symptom descriptions, physical examinations, imaging reports, and treatment plans; the basic information includes age, gender, and abdominal circumference; the symptom descriptions include knee pain levels (0-4) and knee deformity status; the physical examinations include body mass index (BMI) testing, knee tenderness testing, knee range of motion testing, knee stability testing, knee deformity testing, thigh circumference testing, and quadriceps muscle strength testing; the BMI testing is an indicator of whether weight is healthy relative to height; the knee tenderness testing refers to pain induced by applying pressure to specific anatomical locations of the knee joint, and is evaluated according to pain levels; the knee range of motion testing refers to the range of angles of active or passive flexion and extension of the knee joint. The knee joint stability test refers to the knee joint's ability to maintain its normal anatomical position and function under load or stress, assessed through the anterior drawer test, posterior drawer test, Lachman test, varus / valgus stress test, and axis shift test, and evaluated according to stability grading; the knee joint deformity test refers to the degree to which the knee joint deviates from the normal anatomical axis in the coronal plane (frontal view) or sagittal plane (lateral view), measured visually or with a ruler; the thigh circumference test refers to the diameter measured at a fixed distance above the upper edge of the patella (top of the kneecap) using a soft ruler; the quadriceps muscle strength test assesses the contractile strength of the quadriceps (anterior thigh muscle group), including isometric muscle strength in the horizontal and vertical directions of the leg, and the testing methods include manual muscle strength testing, functional testing, and instrumental quantitative testing.

[0027] Furthermore, the natural language text data also includes medical literature, clinical guidelines, and the latest treatment research results related to knee osteoarthritis obtained from authoritative medical databases (such as PubMed, CNKI, etc.) to supplement the latest treatment methods and clinical research progress of knee osteoarthritis and provide data support for subsequent model building.

[0028] Furthermore, the imaging data includes X-ray films, CT images, and MRI images, etc.

[0029] Furthermore, methods for preprocessing natural language text data include: Using Python tools, regular expressions were written with the NLTK and Spacy libraries to remove garbled characters, special symbols, and other interfering information from electronic medical records, thereby ensuring the cleanliness of natural language text data and improving the accuracy of subsequent text processing. The Pandas library was used to structure the natural language text data. By extracting key information, including patient ID, consultation date, and text type (diagnosis record, treatment record), unstructured data in the natural language text data was transformed into structured data, facilitating subsequent analysis and processing. The natural language text data was then categorized and stored according to dimensions such as patient ID, consultation date, and text type before being imported into the database, thus ensuring data traceability and efficiency.

[0030] Furthermore, specific methods for preprocessing radiographic image data include: Based on Python tools, for X-ray images, OpenCV library functions are used to perform histogram equalization (e.g., cv2.createCLAHE), median filtering (e.g., cv2.medianBlur), and affine transformations (e.g., cv2.getAffineTransform and cv2.warpAffine). For CT images, thresholding (e.g., cv2.threshold) is used to remove irrelevant information and improve image quality. For MRI images, denoising (e.g., cv2.fastNlMeansDenoising) and intensity normalization are performed. Three-dimensional volume reconstruction is performed on two-dimensional slices using conventional methods.

[0031] Preferably, the histogram equalization specifically includes: Step a1: Use the grayscale image of the X-ray film as the input image and count the grayscale values ​​of each image. The number of pixels, specifically the formula includes: ; in, express The probability distribution (normalized histogram); express The number of pixels; Indicates the total number of pixels in the input image; This indicates the upper limit of grayscale values ​​(256 for 8-bit images). Step a2: Calculate the cumulative distribution function The specific formulas include: ; in, Indicates the first The probability distribution of each gray value; Step a3: Generate the equalization mapping function, and... Linear mapping to target grayscale range The specific formulas include: ; in, This indicates the output grayscale value (rounded to the nearest integer). Step a4: Convert the grayscale value of each pixel in the input image. Replace with the corresponding This yields an equalized grayscale image.

[0032] Preferably, the specific method for median filtering includes: Step b1: Select the center pixel of the X-ray film. A neighborhood window matrix is ​​constructed using the values ​​of the pixel and its eight neighboring pixels, totaling nine values. The specific expressions include: ; Step b2, will Sort all pixel values ​​in ascending order to obtain an ordered sequence. The specific expressions include: ; in, Indicates the order of pixel values; Step b3, The 5th value in, as median pixel value The specific expressions include: .

[0033] Preferably, the specific method for noise reduction includes: Calculate the first in the MRI image Pixel value after noise reduction of each pixel The specific expressions include: ; in, Represents the set of pixels within the entire MRI image; Indicates the first Pixel values ​​before noise reduction; Indicates the first The pixel and the The weight of each pixel is determined by the similarity between the two pixels, and the specific calculation formula includes: ; in, Indicates the first An image block centered on a pixel; Represents the Gaussian-weighted Euclidean distance; Indicates the Gaussian kernel standard deviation; This represents the filter strength parameter, used to control the attenuation rate. The larger the value, the stronger the noise reduction effect, but the more blurred the details. The normalization factor is represented by the following expressions: .

[0034] Furthermore, the natural language processing method includes: By using a pre-trained GloVe word vector model, words in natural language text data are transformed into word vectors with semantic and contextual relevance. This not only preserves the semantic information of words but also captures the co-occurrence relationships between words, facilitating subsequent model training and inference. Combined with medical professional dictionaries (such as SNOMED CT and MeSH), medical terms in natural language text data are labeled and entity recognized using the BIO (Begin, Inside, Outside) annotation mode. This effectively marks the boundaries of medical terms, enabling the system to identify medical entities such as "knee pain," "joint locking," and "joint swelling" and label their positions in the text. The BIO annotation mode specifically includes: B-SYMPTOM: Marks the starting location of symptoms; I-SYMPTOM: Marks the internal location of symptoms; O: Marks non-physical parts; For example, the text "The patient has knee pain" can be annotated as follows: "Patient O with B-SYMPTOM pain in the knee joint"; For the field of knee joint diseases, the BERT model is fine-tuned and trained (using the TensorFlow or PyTorch framework in Python tools) to adapt to the specific context and medical terminology of the knee joint disease field; through the BERT model, text features related to knee osteoarthritis are extracted from natural language text data.

[0035] Preferably, fine-tuning the training includes: dividing the natural language text data into a training set, a validation set, and a test set according to a preset ratio (e.g., 8:1:1); selecting a cross-validation method; using the cross-entropy loss function as the optimization objective; adjusting the model parameters through the backpropagation algorithm to minimize the prediction error rate; and setting hyperparameters such as the learning rate (e.g., 3e-5, i.e., 0.00003) and batch size (e.g., 16) to ensure training stability and convergence speed.

[0036] In other embodiments, a deep learning model can be used instead of the BERT model to extract more specific text features, such as "pain location", "pain nature", "pain intensity" and "pain duration", so that the system can more comprehensively and thoroughly assess the patient's symptom presentation.

[0037] Furthermore, the convolutional neural network is a U-Net model architecture, specifically including an encoder (downsampling path) and a decoder (upsampling path). The encoder extracts high-level semantic features of the image step by step through convolutional layers and max pooling layers; The decoder improves segmentation accuracy by upsampling high-level semantic features corresponding to the encoder's level.

[0038] Preferably, the decoder includes skip connections to introduce additional operations, optimize information transmission, and thus improve the underutilization of multi-scale information and the vanishing gradient problem in deep networks, such as... Figure 2 As shown, it specifically includes the Inception module (starting module), feature concatenation module, hierarchical attention mechanism module, feature addition module, and feature output module, which are set in sequence. The Inception module extracts input features at different scales through four parallel paths, specifically including: First path: 1×1 convolution, used to capture spatial correlation; The second path: 1×1 convolution + 3×3 convolution, used to extract local detail features; The third path: 1×1 convolution + 5×5 convolution, used to aggregate elements from a wider region; Fourth path: Max pooling + 1×1 convolution, used to obtain high-level semantic features; All paths use the ReLU activation function; The feature splicing module is used to splice the features of the four paths as the output of the Inception module, thereby generating a representation that integrates multi-scale features. The hierarchical attention mechanism module includes spatial attention paths and channel attention paths: The spatial attention path generates spatial attention weights through 1×1 convolution, and performs a dot product operation with the feature map to obtain spatial attention features; The channel attention path generates channel attention weights through global average pooling, and performs channel scaling on the features to obtain channel attention features; The feature addition module is used to add spatial attention features and channel attention features to obtain the feature representation after fusion attention; The feature output module adds the feature representation to the input feature and outputs a residual connection feature, thereby further enhancing the feature representation capability.

[0039] Preferably, the training process of the convolutional neural network includes: (by senior physicians) manually labeling regions of interest in the imaging images as the gold standard using the ITK-SNAP tool; dividing the labeled imaging image data into training and validation sets according to a preset ratio; using classification cross-entropy loss, Dice loss, and boundary loss as loss functions to significantly improve segmentation accuracy; selecting the Adam optimizer; and training the convolutional neural network after setting network parameters such as the initial learning rate (e.g., 3e-4, i.e., 0.0003), batch size (e.g., 32), and number of training epochs (e.g., 100 epochs).

[0040] Furthermore, the specific formula for the classification cross-entropy loss includes: ; in, Indicates the total number of categories; Indicates the category in the label The actual pixel value (takes a value of 0 or 1); This represents the category of the predicted output of the convolutional neural network. The pixel value (takes a value of 0 or 1); The Dice loss is constructed based on the Dice coefficient, which measures the similarity between the predicted segmentation result and the ground truth label. This optimizes model performance by calculating the overlap between the predicted and ground truth regions. The specific formula includes: ; in, This represents the intersection region between the predicted segmentation result and the actual segmentation result; This represents the union region between the predicted segmentation result and the actual segmentation result; and All represent smoothing coefficients, which can take extremely small values ​​(e.g., 1e−7, i.e., 0.0000001) to prevent the denominator from being zero; The boundary loss optimizes the model by constraining the similarity between the predicted segmentation boundary and the true boundary. The core of this loss function is calculating the difference between the gradients of the predicted boundary and the true boundary, specifically expressed as follows: ; in, Indicates the height of the image; Indicates the width of the image; Indicates the first true boundary. Xinggao Di The pixel value of the column width; Indicates the first point at the prediction boundary. Xinggao Di The pixel value of the column width; Represents the gradient operator; This represents the L1 norm, used to calculate the sum of the absolute values ​​of gradient differences.

[0041] Furthermore, the specific methods of the quantitative analysis include: calculating morphological parameter values ​​for morphological features (cartilage, ligaments, bones, etc.) using functions from the Scikit-Image and NumPy libraries; calculating texture parameter values ​​for texture features; and calculating intensity parameter values ​​for intensity features (signal intensity).

[0042] Furthermore, the morphological parameter values ​​specifically include: The hip-knee-ankle angle (HKA) is the lateral angle between the femoral mechanical axis (the line connecting the center of the femoral head and the center of the knee joint) and the tibial mechanical axis (the line connecting the center of the knee joint and the center of the ankle joint). Figure 3 As shown in ∠mHKA; The Lateral Distal Femoral Angle (LDFA) is the angle between the mechanical axis of the femur and the lateral aspect of the distal femoral joint line (the line connecting the lowest point of the medial femoral condyle and the lowest point of the lateral femoral condyle). Figure 4 As shown in ∠LDFA; The medial proximal tibial angle (MPTA) is the angle between the mechanical axis of the tibia and the medial tibial joint line (the line connecting the lowest point of the medial tibial plateau and the lowest point of the lateral tibial plateau). Figure 4 As shown in ∠MPTA; Therefore, the anatomical axis hip-knee angle aHKA = MPTA - LDFA can be calculated to reflect the varus / valgus angle of the knee joint, thereby quantitatively assessing the knee joint deformity status; The femorotibial angle (FTA), also known as the tibial angle, is the medial angle between the femoral anatomical axis (the line connecting the midpoint of the femoral shaft to the center of the knee joint) and the tibial anatomical axis (the line connecting the midpoint of the tibial shaft to the center of the ankle joint). It is used to assess knee joint alignment and varus / valgus angles, thus providing a more precise quantitative assessment of knee joint deformity. Figure 5 As shown; The degree of articular cartilage loss includes the proportion and average thickness of the femoral condyle cartilage and tibial plateau cartilage defects, such as... Figure 6 As shown; The joint space width, i.e., the gap between the femoral condyle and the tibial plateau, can be measured using conventional methods and is used to assess the degree of meniscus wear, such as... Figure 6 As shown; The degree of anterior cruciate ligament tear, i.e., the rate of change in the diameter of the anterior cruciate ligament, such as... Figure 7 As shown.

[0043] Preferably, the specific method for calculating the degree of articular cartilage defect includes: Step c1: Divide the femoral condyle region and tibial plateau region in the segmentation results into two hard bone regions; in each hard bone region, calculate the shortest Euclidean distance from each voxel to the bone boundary in the entire image space. The bone boundary distance field is obtained, and its specific expression includes: ; in, This represents the bone boundary line obtained from boundary detection; Represents the voxel coordinates on the line; Represents the voxel coordinates outside the line; Step c2: Constrain the bone boundary distance field within the corresponding cartilage mask region. The specific expression includes: ; in, This indicates a cartilage indicator function, where the voxel value is 1 within the cartilage mask region and 0 outside the cartilage mask region. Indicates cartilage indication results; Step c3: Extract the cartilage thickness from the cartilage indication results, including the maximum thickness, minimum thickness, and average thickness; calculate the surface area of ​​the indication results as the cartilage region area; calculate the hard bone region area corresponding to the current cartilage region area; use the ratio of the difference between the hard bone region area and the cartilage region area to the hard bone region area as the articular surface cartilage defect ratio, and map it to the ICRS (International Society for Cartilage Repair) classification according to a preset threshold range.

[0044] Preferably, the ICRS classification and preset threshold range include: Grade 0: Normal cartilage; Cartilage defect ratio at the articular surface = 0; Grade 1: Softening and edema are observed on the cartilage surface; 0 < the proportion of articular cartilage defects ≤ 3%; Grade 2: Cartilage shows fissures or fibrosis, but does not reach the full thickness; Grade 3: The proportion of articular surface cartilage defects is ≤10%; Grade 3: Cartilage defects reach the full thickness, but do not involve the bone; 10 < articular surface cartilage defect ratio ≤ 30%; Grade 4: Cartilage defects involve the bone, and even bone exposure occurs; 30% < the proportion of articular surface cartilage defects ≤ 100%.

[0045] Furthermore, the specific method for calculating the degree of anterior cruciate ligament tear includes: Step d1: Perform topology-preserving skeletonization on the binary mask of the anterior cruciate ligament to obtain the centerline of the anterior cruciate ligament. The specific expression includes: ; in, This represents the center line of the anterior cruciate ligament, where the center line pixel is 1 (white) and other areas have pixels of 0 (black). This represents a topology-preserving skeletonization function; This represents a binary mask for the anterior cruciate ligament, where the ligament region has a pixel value of 1 and the background region has a pixel value of 0. Step d2: Calculate the Euclidean distance from the entire image space to the center line of the anterior cruciate ligament. The anterior cruciate ligament distance field is obtained, and its specific expression includes: ; Step d3: Constrain the anterior cruciate ligament distance field within the anterior cruciate ligament binary mask. The specific expression includes: ; in, This indicates the ligament indication result; This represents the ligament indicator function, where the voxel value is 1 within the binary mask region of the anterior cruciate ligament and 0 outside the binary mask region of the anterior cruciate ligament. Step d4: Parametric processing is performed along the centerline of the anterior cruciate ligament (ACL) to obtain several center points; the minimum diameter of the ACL within the cross-section of each center point is calculated; the rate of change of each minimum diameter is statistically analyzed as the degree of ACL tear, and mapped to the knee ligament injury grading according to a preset threshold range; thus, when the rate of change is large, it means that the degree of ACL tear is large, and when the rate of change is small, it means that the degree of ACL tear is small, thereby automatically calculating the degree of ACL tear through the distance field.

[0046] Preferably, the knee ligament injury grading and threshold range include: Grade I injury: minor ligament strain, no obvious rupture; 0 < anterior cruciate ligament tear ≤ 30%; Grade II injury: Partial tear of the ligament with some degree of laxity; 30 < anterior cruciate ligament tear ≤ 90%; Grade III injury: complete ligament rupture, joint instability; 90% < anterior cruciate ligament tear degree ≤ 100%.

[0047] Furthermore, the specific calculation method for the texture parameter values ​​includes: Step e1: Construct the Gray-Level Co-occurrence Matrix (GLCM), which is a probability matrix describing the co-occurrence of gray values ​​in the distance field under specific spatial relationships; perform symmetry and normalization processing on the Gray-Level Co-occurrence Matrix; Step e2: Based on the preset depth layer of the distance field (e.g., it can be divided into surface layer, middle layer, and deep layer), calculate the texture parameter values ​​corresponding to the energy feature, contrast feature, and homogeneity feature under each preset spatial direction using the gray-level co-occurrence matrix: The energy characteristic is used to characterize the uniformity of the distance field (the larger the value, the more uniform the tissue), and the specific calculation formula is as follows: ; in, Represents voxel pairs in the gray-level co-occurrence matrix The normalized probability value; The contrast feature is used to characterize the intensity of local changes (the larger the value, the closer it is to the lesion boundary), and the specific calculation formula is as follows: ; The homogeneity feature is used to characterize local similarity (the larger the value, the smoother the transition), and the specific calculation formula is as follows: ; Step e3: Calculate each preset spatial direction in each preset depth layer. Lower texture features mean and standard deviation After being weighted and summed with the joint space width, it is mapped to the Kellgren-Lawrence grading system (0-4, where 0 is normal and 4 is severe wear) according to a preset threshold range. This better reflects the directionality of collagen and other textures, and can show the lesion evolution gradient corresponding to different depth layers. The specific calculation formula includes: ; ; in, Indicates the number of preset spatial directions.

[0048] Furthermore, the specific calculation method for the intensity parameter value includes: measuring the signal intensity statistical characteristics of the voxel gray value in the IPFP (infrapatellar fat pad) region using MRI images to assess the inflammatory state; the specific calculation formula includes: ; in, The inverse function representing the cumulative signal intensity distribution; This represents the 90th percentile intensity.

[0049] Furthermore, the decision tree model is specifically trained using decision tree algorithms such as C4.5 and CART; feature splitting points are selected based on cost functions such as information gain and Gini index to progressively divide the input features; the decision tree model is evaluated and optimized using performance indicators such as accuracy, recall, and F1 score; and the decision-making process is displayed using visualization methods (nomographs, DCA curves, calibration curves, waterfall plots, etc.) to provide doctors with intuitive decision-making references.

[0050] Preferably, the variables (risk factors) in the decision tree model include: The PO2ELGRISK knee osteoarthritis risk index is used to fuse text features and image quantification features to obtain comprehensive features for overall assessment of knee health; the specific expression is as follows: ; in, Indicates the first Each image quantization feature; Indicates the first Weights of each image quantization feature; This represents the total number of quantized features in the image; Indicates image weight; Indicates the first One text feature; Indicates the first The weights of each text feature; Indicates the total number of text features; Indicates text weight; Age (VO0AGE) is used to highlight the strong correlation between age and knee osteoarthritis (OA) and its importance in decision-making. For example, when age > 60.5 years, the risk of OA is significantly increased. Waist circumference (VO0ABCTRC) is used to highlight the strong correlation between waist circumference and OA and its importance in decision-making. For example, when waist circumference is ≥93.05 (cm), the patient is in a state of central obesity, which is strongly correlated with OA. Body mass index (BMI) is used to highlight the strong correlation between BMI and OA and its importance in decision-making. For example, when BMI is <21.05, it has a protective effect on the knee joint. Isometric leg strength (VOORLHORIZ) is used to highlight the strong correlation and decision-making importance of isometric leg strength on quadriceps femoris (OA). For example, when isometric leg strength is <25.7 (N·m), it may indicate quadriceps atrophy. Isometric leg strength (VOORLVERT) is used to highlight the strong correlation and decision-making importance of isometric leg strength on OA. For example, when isometric leg strength <25.87 (N·m), it can reflect lower limb functional decline. The knee pain status (PO1KSX) is used to highlight the strong correlation between knee pain status and OA and its importance in decision-making. For example, a knee pain status score ≥2.5 may indicate chronic pain. Based on the above variables and preset thresholds, a decision tree model is obtained and visualized, as follows: Figure 8 As shown; where leaf nodes This represents the predicted value of the path sample, used to cumulatively calculate the final risk of a patient developing osteoarthritis. The specific formula includes: ; in, Indicates reaching the th The number of samples in each leaf node; Indicates the first The true label value of a sample (e.g., 0 indicates health, 1 indicates osteoarthritis); This represents the mean of the labels of all samples in the parent node; Indicates the first The set of samples contained in each leaf node; ; in, This indicates the ultimate risk of developing knee osteoarthritis; This indicates the overall prevalence of the training set; Indicates all on the path sum.

[0051] Preferably, the calculation of the knee osteoarthritis risk index further includes: First, the text features and image quantization features are projected and dimensionally aligned respectively, and then the cross-attention energy matrix is ​​calculated. The specific expression includes: ; in, This represents the cross-attention energy matrix, which belongs to the real number field consisting of the number of text features and the number of image quantization features. The matrix elements represent the correlation scores between the text features and the image quantization features. The query vector is obtained by linearly transforming text features. A key vector representing the quantized features of an image; Indicates a unified dimension; Dynamic weights are then generated to achieve automatic weight adjustment, thereby generating more effective comprehensive features. Specific expressions include: ; ; in, express Dynamic weights of image quantization features; Indicates the first The text features of the first... Attention to the quantitative features of an image; This represents the sum of the exponentially weighted attention of all image quantization features; Indicates the first Dynamic weights of text features; Indicates the first The maximum attention span between a text feature and all image quantization features; This represents the sum of the indexed maximum attention across all text features; This indicates a temporary index.

[0052] Furthermore, during the optimization process of the decision tree model, redundant branches can be removed through pruning techniques to improve the simplicity and interpretability of the decision tree model.

[0053] Furthermore, the treatment regimen includes drug therapy and / or physical therapy and / or surgical intervention; The specific strategy for generating the treatment plan is as follows: If the patient's ultimate risk of having knee osteoarthritis is greater than or equal to the first pre-set risk threshold, the treatment option is surgical intervention. If the patient's final risk of having knee osteoarthritis is less than the first preset risk threshold and greater than or equal to the second preset risk threshold, the treatment plan is drug therapy or physical therapy. If the patient's ultimate risk of having knee osteoarthritis is less than the second preset risk threshold, then the treatment plan is physical therapy.

[0054] Furthermore, the generation strategy also includes a surgical plan determination method: Step f1: Deconstruct surgical parameters from comprehensive characteristics, specifically including pain location, joint space width, degree of anterior cruciate ligament tear, varus / valgus angle, and age; Step f2: Determine the surgical parameters using the surgical instruction function. Specific determination conditions include: Pain location indicator value: If the pain is located on the inside of the knee joint, the value is 1; otherwise, it is 0. Joint space width indicator: If ≤2 mm, the value is 1; otherwise, it is 0. Anterior cruciate ligament tear severity indicator: 1 for grade III injury, 0 otherwise; Inward / outward folding angle indicator value: If >10°, the value is 1; otherwise, it is 0. Age indicator: 1 if <60 years old, 0 otherwise; Step f3: Generate a surgical plan using a surgical generation strategy, specifically including: If the joint space width indicator value is 1 and the pain location indicator value is 0, then TKA (total knee replacement) surgery is recommended. If the varus / valgus angle is 1 and the age indicator is 1, then HTO (high tibial osteotomy) / DFO (distal femoral osteotomy) surgery is recommended. If the anterior cruciate ligament tear severity indicator value is 0, UKA (unicompartmental replacement) surgery is recommended.

[0055] Example 2: This embodiment provides a decision tree device for the treatment of knee osteoarthritis, including a processor, a memory, and a bus. The memory stores instructions and data read by the processor, and the processor is used to call the instructions and data in the memory to realize the knee osteoarthritis treatment decision tree system as described above. The bus connects the various functional components for information transmission.

[0056] In another embodiment, this solution can also be implemented using an integrated device, which may include corresponding modules that perform one or more steps in the various embodiments described above. A module may be one or more hardware modules specifically configured to perform the corresponding step, or implemented by a processor configured to perform the corresponding step, or stored in a computer-readable medium for implementation by a processor, or implemented through some combination thereof.

[0057] The processor executes the various methods and processes described above. For example, the method implementations in this scheme can be implemented as software programs tangibly contained in a machine-readable medium, such as memory. In some implementations, part or all of the software program can be loaded and / or installed via memory and / or a communication interface. When the software program is loaded into memory and executed by the processor, one or more steps of the methods described above can be performed. Alternatively, in other implementations, the processor can be configured to execute one of the methods described above by any other suitable means (e.g., by means of firmware).

[0058] This device can be implemented using a bus architecture. A bus architecture can include any number of interconnect buses and bridges, depending on the specific application of the hardware and overall design constraints. The bus connects various circuits, including one or more processors, memory, and / or hardware modules. The bus can also connect various other circuits such as peripherals, voltage regulators, power management circuitry, external antennas, etc.

[0059] Buses can be Industry Standard Architecture (ISA) buses, Peripheral Component Interconnect (PCI) buses, or Extended Industry Standard Component (EISA) buses, etc. Buses can be divided into address buses, data buses, control buses, etc.

[0060] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A decision tree system for the treatment of knee osteoarthritis, characterized in that, It includes a data receiving module, a data processing module, and a result generation module; The data receiving module is used to receive the patient's natural language text data and imaging image data, and to preprocess both of them respectively; The data processing module includes a text feature unit, an image feature unit, and a decision unit; The text feature unit extracts text features related to knee osteoarthritis from natural language text data using natural language processing methods; and performs structured processing on the text features. The image feature unit identifies and segments the region of interest in the radiographic image data through a convolutional neural network to obtain the segmentation result; based on the segmentation result, it extracts morphological features, texture features and intensity features related to knee osteoarthritis and then performs quantitative analysis to obtain image quantitative features. The decision-making unit fuses text features and image quantization features to obtain comprehensive features; Based on comprehensive features, a treatment plan for knee osteoarthritis is output through a decision tree model; The result generation module is used to distribute the treatment plan externally.

2. The system according to claim 1, characterized in that, The natural language text data includes electronic medical records, doctor's records, and medical history questionnaires; the electronic medical records include the patient's basic information, symptom descriptions, physical examinations, imaging reports, and treatment plans; the basic information includes age and gender; the symptom descriptions include knee pain levels and knee deformity status; the physical examinations include body mass index (BMI) testing, knee tenderness testing, knee range of motion testing, knee stability testing, knee deformity testing, thigh circumference testing, and quadriceps muscle strength testing.

3. The system according to claim 1, characterized in that, The imaging data includes X-ray films, CT images, and MRI images.

4. The system according to claim 1, characterized in that, The convolutional neural network is a U-Net model architecture, specifically including an encoder and a decoder; The encoder extracts high-level semantic features of the image step by step through convolutional layers and max pooling layers; The decoder improves segmentation accuracy by upsampling high-level semantic features corresponding to the encoder's level.

5. The system according to claim 4, characterized in that, The decoder includes skip connections, specifically comprising an Inception module, a feature concatenation module, a hierarchical attention mechanism module, a feature addition module, and a feature output module arranged sequentially. The Inception module extracts input features at different scales through four parallel paths, specifically including: First path: 1×1 convolution; Second path: 1×1 convolution + 3×3 convolution; Third path: 1×1 convolution + 5×5 convolution; Fourth path: Max pooling + 1×1 convolution; All paths use the ReLU activation function; The feature splicing module is used to splice the features of the four paths as the output of the Inception module; The hierarchical attention mechanism module includes spatial attention paths and channel attention paths: The spatial attention path generates spatial attention weights through 1×1 convolution, and performs a dot product operation with the feature map to obtain spatial attention features; The channel attention path generates channel attention weights through global average pooling, and performs channel scaling on the features to obtain channel attention features; The feature addition module is used to add spatial attention features and channel attention features to obtain the feature representation after fusion attention; The feature output module adds the feature representation to the input feature and outputs a residual connection feature.

6. The system according to claim 5, characterized in that, The training process of a convolutional neural network includes: using the ITK-SNAP tool to annotate regions of interest in the image as the gold standard; dividing the annotated image data into training and validation sets according to a preset ratio; using classification cross-entropy loss, Dice loss, and boundary loss as loss functions; selecting the Adam optimizer; and training the convolutional neural network after setting the initial learning rate, batch size, and number of training epochs.

7. The system according to claim 6, characterized in that, The specific formula for the classification cross-entropy loss includes: ； in, Indicates the total number of categories; Indicates the category in the label The actual pixel values; This represents the category of the predicted output of the convolutional neural network. Pixel values; The specific formula for the Dice loss includes: ； in, This represents the intersection region between the predicted segmentation result and the actual segmentation result; This represents the union region between the predicted segmentation result and the actual segmentation result; and All represent smoothing coefficients; The specific expression for the boundary loss includes: ； in, Indicates the height of the image; Indicates the width of the image; Indicates the first true boundary. Xinggao Di The pixel value of the column width; Indicates the first point at the prediction boundary. Xinggao Di The pixel value of the column width; Represents the gradient operator; This represents the L1 norm, used to calculate the sum of the absolute values ​​of gradient differences.

8. The system according to claim 1, characterized in that, The specific methods for quantitative analysis include: calculating morphological parameter values ​​for morphological features, calculating texture parameter values ​​for texture features, and calculating intensity parameter values ​​for intensity features using functions from the Scikit-Image and NumPy libraries.

9. The system according to claim 8, characterized in that, The morphological parameter values ​​specifically include: Hip-knee-ankle angle (HKA); LDFA of the distal lateral angle of the femur; MPTA of the proximal medial tibial angle; Anatomical axis hip-knee angle aHKA = MPTA - LDFA; Femoral-tibia angle FTA; The proportion of articular cartilage defects, including the proportion and average thickness of defects in the femoral condyle cartilage and tibial plateau cartilage; Joint space width; The degree of anterior cruciate ligament tear, i.e. the rate of change in the diameter of the anterior cruciate ligament.

10. A decision tree device for the treatment of knee osteoarthritis, characterized in that, The system includes a processor, a memory, and a bus. The memory stores instructions and data read by the processor, and the processor is used to call the instructions and data in the memory to implement the system as described in any one of claims 1-9. The bus connects the functional components for transmitting information.

Images