A joint cavity positioning method and device based on deep learning model

Abstract

The present application relates to the technical field of joint cavity positioning, and specifically provides a joint cavity positioning method and device based on a deep learning model, the method comprising the steps of: S1, acquiring an arthroscopic image, and preprocessing the arthroscopic image to obtain an enhanced image; S2, acquiring joint cavity position information based on a pre-trained joint cavity positioning model and the enhanced image; the pre-training of the joint cavity positioning model comprises: A1, training a pre-built deep learning model using a pre-collected general medical image dataset to obtain a preliminary positioning model; A2, expanding a pre-collected arthroscopic image dataset of rare bone and joint diseases to obtain an expanded image dataset, and then dynamically adjusting the regularization strength of the preliminary positioning model based on an adaptive regularization coefficient function according to the joint cavity morphological diversity index of the expanded image dataset to obtain a joint cavity positioning model; the method can accurately locate the joint cavity of rare bone and joint diseases.

Description

Technical Field

[0001] The present application relates to the field of joint cavity positioning technology, and specifically, to a joint cavity positioning method and device based on a deep learning model. Background Art

[0002] Human joints are complex, surrounded by numerous blood vessels, nerves, tendons, and ligaments. Locating the joint cavity typically requires repeated puncture attempts and gradual corrections. This traditional surgical approach is not only time-consuming but also prone to causing unnecessary damage to human tissue. To shorten surgical time and minimize damage to human tissue, existing technologies are exploring the use of joint cavity localization models derived from deep learning models to locate the joint cavity.

[0003] However, in specialized medical centers specializing in rare bone and joint diseases, applying the aforementioned joint cavity localization technology to the localization of the joint cavities of rare bone and joint diseases presents unprecedented challenges. This is because rare bone and joint diseases have a low incidence rate, which means that the amount of image data corresponding to these diseases that can be used for model training is severely insufficient. Deep learning models are data-driven models, meaning they rely on large amounts of image data for training to ensure that the models can accurately identify and locate joint cavities. Therefore, existing technologies have the problem that due to the insufficient amount of training data corresponding to rare bone and joint diseases, the joint cavity localization models derived from deep learning models are unable to accurately locate the joint cavities of rare bone and joint diseases.

[0004] Furthermore, the pathological features of rare bone and joint diseases often exhibit a high degree of heterogeneity. Joint cavity morphology can vary significantly between patients, or within the same patient at different stages of the disease course. This diversity in joint cavity morphology makes it difficult for deep learning-based joint cavity localization models to learn universal features, resulting in a decrease in the generalization and positioning accuracy of the joint cavity localization model. Existing technologies still suffer from the problem of reduced generalization and positioning accuracy due to the diversity of joint cavity morphology.

[0005] There is no effective technical solution to the above problems. It should be noted that the above information disclosed in this section is only used to understand the background of the present invention, and therefore may contain information that does not constitute prior art. Summary of the Invention

[0006] The purpose of this application is to provide a joint cavity positioning method and device based on a deep learning model, which can effectively solve the problem that the joint cavity positioning model obtained based on the deep learning model cannot accurately locate the joint cavity of rare bone and joint diseases due to insufficient training data corresponding to rare bone and joint diseases, and the generalization ability and positioning accuracy of the joint cavity positioning model are reduced due to the diversity of joint cavity morphology.

[0007] In a first aspect, the present application provides a joint cavity positioning method based on a deep learning model for positioning the joint cavity of rare bone and joint diseases. The joint cavity positioning method based on a deep learning model comprises the following steps:

[0008] S1. Acquire an arthroscopic image and preprocess the arthroscopic image to obtain an enhanced image, wherein the preprocessing includes contrast enhancement;

[0009] S2. Acquire joint cavity position information based on a pre-trained joint cavity positioning model and the enhanced image. The joint cavity positioning model is used to locate the joint cavity in the enhanced image and output the joint cavity position.

[0010] Pre-training of the joint cavity positioning model includes:

[0011] A1. Use a pre-collected general medical image dataset to train a pre-built deep learning model to obtain a preliminary positioning model.

[0012] A2. Expand the pre-collected arthroscopic image dataset of rare bone and joint diseases to obtain an expanded image dataset. Then, based on the adaptive regularization coefficient function, dynamically adjust the regularization strength of the preliminary positioning model according to the joint cavity morphology diversity index of the expanded image dataset to obtain a joint cavity positioning model.

[0013] The present application provides a joint cavity positioning method based on a deep learning model, which is equivalent to first training a pre-constructed deep learning model with a pre-collected general medical image data set so that the deep learning model learns the general features of medical images and the basic knowledge of joint cavity positioning, and then dynamically adjusts the regularization strength of the preliminary positioning model according to the joint cavity morphology diversity index of the expanded image data set based on an adaptive regularization coefficient function so that the deep learning model learns the unique features of the joint cavity of rare bone and joint diseases while avoiding overfitting. Therefore, the joint cavity positioning model of the present application has good generalization ability and positioning accuracy, and the joint cavity positioning model can more accurately locate the joint cavity of rare bone and joint diseases. That is, the present application can effectively solve the problem that the joint cavity positioning model obtained based on the deep learning model cannot accurately locate the joint cavity of rare bone and joint diseases due to insufficient training data corresponding to rare bone and joint diseases, and the generalization ability and positioning accuracy of the joint cavity positioning model are reduced due to the diversity of joint cavity morphology.

[0014] Optionally, step A2 includes:

[0015] A21. Expanding a pre-collected rare bone and joint disease arthroscopic image dataset by randomly rotating, scaling, and flipping each image in the pre-collected rare bone and joint disease arthroscopic image dataset to obtain an expanded image dataset, where the expanded image dataset includes a plurality of expanded images;

[0016] A22. Obtain an image quality score for each expanded image using a pre-built image quality assessment model to obtain multiple image quality scores;

[0017] A23. Normalize all image quality scores to the interval [0, 1] to obtain the quality weight corresponding to each expanded image, and obtain the joint cavity morphological feature vector corresponding to each expanded image based on the local binary pattern histogram corresponding to the expanded image;

[0018] A24. Perform weighted average calculation based on all mass weights and all joint cavity morphology feature vectors to obtain a joint cavity morphology diversity index;

[0019] A25. Based on the adaptive regularization coefficient function, the regularization strength of the preliminary positioning model is dynamically adjusted according to the joint cavity morphology diversity index of the expanded image data set to obtain the joint cavity positioning model.

[0020] Since this technical solution is equivalent to introducing image quality evaluation and joint cavity morphology diversity indicators in the data expansion process, and using the joint cavity morphology diversity indicators to dynamically adjust the regularization strength of model training, this technical solution can effectively solve the problem of low accuracy of joint cavity morphology diversity caused by uneven quality of expanded images, thereby effectively improving the generalization ability and positioning accuracy of the joint cavity positioning model.

[0021] Optionally, the image quality assessment model is used to extract clarity features, contrast features, and noise level features of each expanded image, and output corresponding clarity scores, contrast scores, and noise scores. Step A22 includes:

[0022] A221. For each expanded image, calculate its corresponding image quality score based on its corresponding clarity score, contrast score, and noise score based on a weighted fusion algorithm.

[0023] Since this technical solution can first use the image quality assessment model to independently evaluate its clarity, contrast and noise level and output the corresponding score, and then calculate the image quality score corresponding to each expanded image based on the clarity score, contrast score and noise score based on the weighted fusion algorithm, it avoids the situation where the image quality score cannot accurately reflect the overall quality of the image due to evaluating the quality of the expanded image based on only one dimension, thereby effectively improving the training efficiency and positioning accuracy of the preliminary positioning model, and then effectively improving the generalization ability and positioning accuracy of the joint cavity positioning model.

[0024] Optionally, step A221 includes:

[0025] A2211. For each expanded image, divide the expanded image into a plurality of non-overlapping image blocks;

[0026] A2212. Obtaining gradient distribution features, color histogram features, and texture features of each image block. The gradient distribution features are obtained by calculating the gradient magnitude and direction of each image block. The color histogram features are obtained by statistically analyzing the color distribution of each image block in the HSV color space. The texture features are obtained by calculating the gray-level co-occurrence matrix of each image block.

[0027] A2213. Calculating the clarity confidence, contrast confidence, and noise confidence of each image block based on the gradient distribution features, color histogram features, and texture features, and normalizing the clarity confidence, contrast confidence, and noise confidence based on the Aoftmax function to obtain the clarity weight, contrast weight, and noise weight of the weighted fusion algorithm;

[0028] A2214. Calculate the corresponding image quality score based on the corresponding clarity score, contrast score, and noise score based on the weighted fusion algorithm.

[0029] This technical solution is equivalent to a method of adaptively determining weights based on the local features of different areas in the expanded image. Therefore, this technical solution can make the image quality score obtained based on the weighted fusion algorithm more refined, thereby improving the accuracy of the image quality score, and further improving the training effect and positioning accuracy of the joint cavity positioning model.

[0030] Optionally, step A2211 includes:

[0031] A22111. For each expanded image, calculating a gradient magnitude image of the expanded image;

[0032] A22112, selecting the pixel with the largest gradient amplitude in the gradient amplitude image as the seed point;

[0033] A22113, expanding toward the neighborhood with the seed point as the center, and merging pixels in the neighborhood whose gradient magnitude difference with the seed point is less than a preset threshold into the current image block, until no pixel whose gradient magnitude difference with the seed point is less than the preset threshold exists in the gradient magnitude image;

[0034] A22114. Analyze whether there are undivided pixels in the gradient amplitude image. If so, select the pixel with the largest gradient amplitude among the undivided pixels as the new seed point and return to step A22113. If not, execute step 2212.

[0035] Since this technical solution can merge pixels with similar gradient amplitudes into the same image block, that is, the gradient features of pixels within the same image block are similar, this technical solution can effectively improve the uniformity of image features within the image block, thereby effectively further improving the accuracy of image quality scoring.

[0036] Optionally, step A22111 includes:

[0037] A221111. For each expanded image, convert the expanded image to the HSV color space and perform blurring on the V channel of the HSV color space to obtain a light intensity map.

[0038] A221112. Calculate the illumination compensation coefficient corresponding to each pixel in the expanded image according to the illumination intensity map;

[0039] A221113. Adjust the RGB value corresponding to each pixel in the expanded image according to the illumination compensation coefficient;

[0040] A221114. Calculate the gradient magnitude image of the expanded image.

[0041] Before calculating the gradient amplitude image of the expanded image, this technical solution first obtains the illumination intensity map corresponding to the expanded image and adjusts the RGB values ​​of the pixels in the expanded image based on the illumination intensity map to perform illumination balancing on the expanded image. Therefore, this technical solution can effectively reduce the impact of uneven illumination on the calculation of the gradient amplitude image, thereby effectively improving the accuracy of the gradient amplitude image.

[0042] Optionally, step A25 includes:

[0043] A251. Obtaining a high-frequency sub-band energy intensity feature corresponding to each expanded image based on a wavelet transform, and obtaining a loss weight corresponding to each expanded image according to the high-frequency sub-band energy intensity feature and a first preset conversion relationship;

[0044] A252. Based on the adaptive regularization coefficient function, the regularization strength of the preliminary positioning model is dynamically adjusted according to the joint cavity morphology diversity index of the expanded image dataset. During the adjustment process, the Adam optimizer is used to minimize the cross entropy loss function of the preliminary positioning model by multiplying the cross entropy loss value of each expanded image by the corresponding loss weight to obtain the joint cavity positioning model.

[0045] Optionally, step S2 includes:

[0046] S21, the feature points of the current frame image are matched with the feature points of the previous frame image to obtain a matching pair of feature points;

[0047] S22. Estimate the homography matrix of the current frame image relative to the previous frame image based on the matched feature point pairs using the RANSAC algorithm. The homography matrix describes the transformation relationship of the arthroscopic view:

[0048] S23, decomposing the homography matrix into a rotation matrix and a translation vector, and performing rotation and translation operations on the convolution kernel of the pre-trained joint cavity positioning model according to the rotation matrix and the translation vector, so that the joint cavity positioning model adapts to the current perspective;

[0049] S24. Input the enhanced image into the joint cavity positioning model to obtain joint cavity position information.

[0050] Optionally, the pre-processing further includes edge sharpening.

[0051] In a second aspect, the present application further provides a joint cavity positioning device based on a deep learning model for positioning the joint cavity of rare bone and joint diseases. The joint cavity positioning device based on a deep learning model comprises the following steps:

[0052] An enhanced image acquisition module is used to acquire an arthroscopic image and preprocess the arthroscopic image to obtain an enhanced image, wherein the preprocessing includes contrast enhancement;

[0053] The joint cavity position information acquisition module acquires the joint cavity position information based on the pre-trained joint cavity positioning model and the enhanced image. The joint cavity positioning model is used to locate the joint cavity in the enhanced image and output the joint cavity position.

[0054] Pre-training of the joint cavity positioning model includes:

[0055] A1. Use a pre-collected general medical image dataset to train a pre-built deep learning model to obtain a preliminary positioning model.

[0056] A2. Expand the pre-collected arthroscopic image dataset of rare bone and joint diseases to obtain an expanded image dataset. Then, based on the adaptive regularization coefficient function, dynamically adjust the regularization strength of the preliminary positioning model according to the joint cavity morphology diversity index of the expanded image dataset to obtain a joint cavity positioning model.

[0057] The present application provides a joint cavity positioning device based on a deep learning model, which is equivalent to first training a pre-constructed deep learning model through a pre-collected general medical image data set so that the deep learning model learns the general features of medical images and the basic knowledge of joint cavity positioning, and then dynamically adjusts the regularization strength of the preliminary positioning model according to the joint cavity morphology diversity index of the expanded image data set based on an adaptive regularization coefficient function so that the deep learning model learns the unique features of the joint cavity of rare bone and joint diseases while avoiding overfitting. Therefore, the joint cavity positioning model of the present application has good generalization ability and positioning accuracy, and the joint cavity positioning model can more accurately locate the joint cavity of rare bone and joint diseases. That is, the present application can effectively solve the problem that the joint cavity positioning model obtained based on the deep learning model cannot accurately locate the joint cavity of rare bone and joint diseases due to insufficient training data corresponding to rare bone and joint diseases, and the generalization ability and positioning accuracy of the joint cavity positioning model are reduced due to the diversity of joint cavity morphology.

[0058] From the above, it can be seen that the joint cavity positioning method and device based on a deep learning model provided by the present application is equivalent to first training a pre-constructed deep learning model through a pre-collected general medical image data set so that the deep learning model learns the general features of medical images and the basic knowledge of joint cavity positioning, and then dynamically adjusts the regularization strength of the preliminary positioning model according to the joint cavity morphology diversity index of the expanded image data set based on an adaptive regularization coefficient function so that the deep learning model learns the unique features of the joint cavity of rare bone and joint diseases while avoiding overfitting. Therefore, the joint cavity positioning model of the present application has good generalization ability and positioning accuracy, and the joint cavity positioning model can more accurately locate the joint cavity of rare bone and joint diseases, that is, the present application can effectively solve the problem that the joint cavity positioning model obtained based on the deep learning model cannot accurately locate the joint cavity of rare bone and joint diseases due to insufficient training data corresponding to rare bone and joint diseases, and the generalization ability and positioning accuracy of the joint cavity positioning model are reduced due to the diversity of joint cavity morphology. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 A flowchart of a joint cavity positioning method based on a deep learning model provided in an embodiment of the present application.

[0060] Figure 2 A schematic structural diagram of a joint cavity positioning device based on a deep learning model provided in an embodiment of the present application.

[0061] Reference numerals: 1. Enhanced image acquisition module; 2. Joint cavity position information acquisition module. DETAILED DESCRIPTION

[0062] The technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all of the embodiments. The components of the embodiments of the present application generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the application for protection, but merely represents the selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without making creative work fall within the scope of protection of the present application.

[0063] It should be noted that similar reference numerals and letters represent similar items in the following drawings. Therefore, once an item is defined in one drawing, it does not need to be further defined or explained in subsequent drawings. At the same time, in the description of this application, the terms "first", "second", etc. are only used to distinguish the description and should not be understood as indicating or implying relative importance.

[0064] First, as Figure 1 As shown, the present application provides a joint cavity positioning method based on a deep learning model for positioning the joint cavity of rare bone and joint diseases. The joint cavity positioning method based on a deep learning model includes the following steps:

[0065] S1. Acquire an arthroscopic image and preprocess the arthroscopic image to obtain an enhanced image, wherein the preprocessing includes contrast enhancement;

[0066] S2. Acquire joint cavity position information based on a pre-trained joint cavity positioning model and the enhanced image. The joint cavity positioning model is used to locate the joint cavity in the enhanced image and output the joint cavity position.

[0067] Pre-training of the joint cavity positioning model includes:

[0068] A1. Use a pre-collected general medical image dataset to train a pre-built deep learning model to obtain a preliminary positioning model.

[0069] A2. Expand the pre-collected arthroscopic image dataset of rare bone and joint diseases to obtain an expanded image dataset. Then, based on the adaptive regularization coefficient function, dynamically adjust the regularization strength of the preliminary positioning model according to the joint cavity morphology diversity index of the expanded image dataset to obtain a joint cavity positioning model.

[0070] The preprocessing operation in step S1 is intended to improve the quality of the arthroscopic image. The preprocessing includes contrast enhancement (such as histogram equalization or contrast-limited adaptive histogram equalization algorithm). The contrast enhancement is used to enhance the contrast of the arthroscopic image to make the boundary between the joint cavity and the surrounding tissue clearer and provide higher quality input data for the subsequent joint cavity positioning model, thereby effectively improving the accuracy of joint cavity positioning.

[0071] In step S2, the pre-trained joint cavity localization model is the core of realizing automatic joint cavity localization. The model can receive the pre-processed enhanced image as input and output the location information of the joint cavity (such as the bounding box or segmentation mask of the joint cavity in the enhanced image).

[0072] The pre-training process of the joint cavity localization model is divided into steps A1 and A2. In step A1, a pre-constructed deep learning model is first trained using a pre-collected general medical image dataset to obtain a preliminary localization model. The general medical image dataset may include various medical images corresponding to common bone and joint diseases (such as CT images and MRI images). This embodiment enables the deep learning model to learn the general features of medical images and basic knowledge of joint cavity localization by using the pre-collected general medical image dataset (equivalent to pre-training the deep learning model using general data). Even when the amount of data on rare bone and joint diseases is insufficient, the model still has a certain degree of joint cavity localization capability. Because rare bone and joint diseases have a low incidence rate, that is, the pre-collected arthroscopic image dataset of rare bone and joint diseases contains a small amount of data. Therefore, step A2 can expand the arthroscopic image dataset of rare bone and joint diseases by using existing data expansion methods to increase the amount of image data corresponding to rare bone and joint diseases that can be used for model training and increase the diversity of image data corresponding to rare bone and joint diseases used for model training, so that the joint cavity localization model has higher generalization ability. Then, the joint cavity morphology diversity index of the expanded image data set is calculated. This index is used to evaluate the degree of change in the joint cavity morphology in the expanded data set. The joint cavity morphology diversity index can be calculated based on the morphological features of the image. Afterwards, based on the adaptive regularization coefficient function, the regularization strength of the preliminary positioning model is dynamically adjusted according to the joint cavity morphology diversity index. Regularization is a commonly used technology to prevent model overfitting. This embodiment can limit the complexity of the model by adding a regularization term to the loss function of the deep learning model. The adaptive regularization coefficient function of this embodiment can dynamically adjust the regularization strength according to the diversity of joint cavity morphology. Specifically, when the diversity of joint cavity morphology is high, this embodiment increases the regularization strength to prevent the model from overfitting to a joint cavity of a specific morphology, thereby effectively improving the generalization ability of the model; when the diversity of joint cavity morphology is low, this embodiment reduces the regularization strength so that the model can better learn the unique characteristics of the joint cavity of rare bone and joint diseases. Therefore, the joint cavity positioning model obtained after training in step A2 can more accurately locate the joint cavity of rare bone and joint diseases. It should be understood that this embodiment is equivalent to first using a general medical image dataset to train the model to obtain a general joint cavity positioning model (equivalent to a general model applicable to common bone and joint diseases), and then using the expanded arthroscopic image dataset of rare bone and joint diseases to train the general joint cavity positioning model to obtain a joint cavity positioning model applicable to rare bone and joint diseases.

[0073] The present application provides a joint cavity positioning method based on a deep learning model, which is equivalent to first training a pre-constructed deep learning model with a pre-collected general medical image data set so that the deep learning model learns the general features of medical images and the basic knowledge of joint cavity positioning, and then dynamically adjusts the regularization strength of the preliminary positioning model according to the joint cavity morphology diversity index of the expanded image data set based on an adaptive regularization coefficient function so that the deep learning model learns the unique features of the joint cavity of rare bone and joint diseases while avoiding overfitting. Therefore, the joint cavity positioning model of the present application has good generalization ability and positioning accuracy, and the joint cavity positioning model can more accurately locate the joint cavity of rare bone and joint diseases. That is, the present application can effectively solve the problem that the joint cavity positioning model obtained based on the deep learning model cannot accurately locate the joint cavity of rare bone and joint diseases due to insufficient training data corresponding to rare bone and joint diseases, and the generalization ability and positioning accuracy of the joint cavity positioning model are reduced due to the diversity of joint cavity morphology.

[0074] In some preferred embodiments, step A2 comprises:

[0075] A21. Expanding a pre-collected rare bone and joint disease arthroscopic image dataset by randomly rotating, scaling, and flipping each image in the pre-collected rare bone and joint disease arthroscopic image dataset to obtain an expanded image dataset, where the expanded image dataset includes a plurality of expanded images;

[0076] A22. Obtain an image quality score for each expanded image using a pre-built image quality assessment model to obtain multiple image quality scores;

[0077] A23. Normalize all image quality scores to the interval [0, 1] to obtain the quality weight corresponding to each expanded image, and obtain the joint cavity morphological feature vector corresponding to each expanded image based on the local binary pattern histogram corresponding to the expanded image;

[0078] A24. Perform weighted average calculation based on all mass weights and all joint cavity morphology feature vectors to obtain a joint cavity morphology diversity index;

[0079] A25. Based on the adaptive regularization coefficient function, the regularization strength of the preliminary positioning model is dynamically adjusted according to the joint cavity morphology diversity index of the expanded image data set to obtain the joint cavity positioning model.

[0080] Step A21 can achieve data enhancement for a pre-collected arthroscopic image dataset of rare bone and joint diseases by randomly rotating, scaling, and flipping each image in the pre-collected arthroscopic image dataset of rare bone and joint diseases, thereby increasing the diversity of the training data. In step A22, the image quality assessment model can be a pre-trained convolutional neural network, which is configured to receive image input and output a quality score. Step A23 can linearly map all quality scores to the [0,1] interval by utilizing the existing minimum-maximum normalization method. The quality weight of step A23 is directly determined by the normalized quality score. The joint cavity morphology feature vector is obtained by calculating the local binary pattern histogram of each expanded image. The local binary pattern histogram can capture the local texture information of the image and serve as a representation of the joint cavity morphology. The joint cavity morphology diversity index of step A24 is obtained by weighted averaging. This index can comprehensively reflect the image quality and joint cavity morphology distribution of the expanded data set. In the weighted averaging, each joint cavity morphology feature vector is weighted by its corresponding quality weight. The higher the quality weight, the greater the contribution of the corresponding image in the calculation of the diversity index. Specifically, when the joint cavity morphology diversity index is high, it indicates that the joint cavity morphology of the data set is relatively different. At this time, increasing the regularization strength can improve the generalization ability of the model and avoid overfitting the model to a joint cavity of a specific morphology. Conversely, when the joint cavity morphology diversity index is low, the regularization strength can be appropriately reduced to enable the model to better learn the subtle features in the data. Since this embodiment is equivalent to introducing image quality assessment and joint cavity morphology diversity index in the data expansion process, and using the joint cavity morphology diversity index to dynamically adjust the regularization strength of the model training, this embodiment can effectively solve the problem of low accuracy of joint cavity morphology diversity due to uneven quality of the expanded images, thereby effectively improving the generalization ability and positioning accuracy of the joint cavity positioning model.

[0081] In some preferred embodiments, the image quality assessment model is used to extract the clarity features, contrast features, and noise level features of each expanded image and output the corresponding clarity score, contrast score, and noise score. Step A22 includes:

[0082] A221. For each expanded image, calculate its corresponding image quality score based on its corresponding clarity score, contrast score, and noise score based on a weighted fusion algorithm.

[0083] The image quality assessment model of this embodiment is configured to analyze three key quality dimensions of the expanded image: clarity, contrast, and noise level. This embodiment can extract clarity features based on image gradient information, for example, by calculating the Laplacian response or gradient amplitude of the image. The higher the response value or amplitude, the clearer the image. This embodiment can extract contrast features based on the image's grayscale distribution, for example, by calculating the standard deviation or information entropy of the image's grayscale values. The larger the standard deviation or the higher the information entropy, the higher the image contrast. This embodiment can extract noise level features by analyzing the image's frequency domain characteristics or fluctuations in local pixel values. For example, by calculating the energy of the image's high-frequency components or the variance of pixel values ​​in a local region. The lower the energy or the smaller the variance, the lower the image noise level. After obtaining the clarity, contrast, and noise level features, the image quality assessment model converts these features into corresponding scores. The score conversion can be performed using linear mapping, nonlinear mapping, or table lookup. In step A221, a weighted fusion algorithm is used to integrate the clarity score, contrast score, and noise score to obtain a final image quality score. The specific implementation of the weighted fusion algorithm can be varied. For example, a weight can be assigned to each score, and then the weighted average is calculated as the image quality score. The weight can be set based on experience or obtained through learning. Since this embodiment can first use the image quality assessment model to independently assess its clarity, contrast and noise level, and output the corresponding score, and then calculate the image quality score corresponding to each expanded image based on the clarity score, contrast score and noise score based on the weighted fusion algorithm, it avoids the situation where the image quality score cannot accurately reflect the overall quality of the image due to the quality of the expanded image being assessed based on only one dimension, thereby effectively improving the training efficiency and positioning accuracy of the preliminary positioning model, and then effectively improving the generalization ability and positioning accuracy of the joint cavity positioning model. Specifically, the formula of the weighted fusion algorithm of this embodiment is: ; Where Q represents the image quality score, w c represents the clarity weight, C represents the clarity score, w o represents contrast weight, O represents contrast score, w n represents the noise weight, and N represents the noise score.

[0084] In some preferred embodiments, step A221 includes:

[0085] A2211. For each expanded image, divide the expanded image into a plurality of non-overlapping image blocks;

[0086] A2212. Obtaining gradient distribution features, color histogram features, and texture features of each image block. The gradient distribution features are obtained by calculating the gradient magnitude and direction of each image block. The color histogram features are obtained by statistically analyzing the color distribution of each image block in the HSV color space. The texture features are obtained by calculating the gray-level co-occurrence matrix of each image block.

[0087] A2213. Calculating the clarity confidence, contrast confidence, and noise confidence of each image block based on the gradient distribution features, color histogram features, and texture features, and normalizing the clarity confidence, contrast confidence, and noise confidence based on the Aoftmax function to obtain the clarity weight, contrast weight, and noise weight of the weighted fusion algorithm;

[0088] A2214. Calculate the corresponding image quality score based on the corresponding clarity score, contrast score, and noise score based on the weighted fusion algorithm.

[0089] Step A2211 can divide the expanded image into multiple non-overlapping image blocks by partitioning the expanded image using a fixed-size grid. This embodiment can obtain gradient distribution features by calculating the gradient magnitude and direction of the image block using the Sobel operator or the Canny operator. This embodiment can obtain color histogram features by statistically analyzing the pixel distribution of each color channel of the image block in the HSV color space. This embodiment can obtain texture features by calculating statistics such as the energy, contrast, correlation, and entropy of the gray-level co-occurrence matrix, which can characterize the texture information of the image block. The clarity confidence, contrast confidence, and noise confidence in step A2213 can be obtained using pre-trained classifiers or regression models, respectively. These models use the gradient distribution features, color histogram features, and texture features extracted in step A2212 as input and output corresponding confidence scores. The Aoftmax function is used to normalize the confidence scores to more appropriately weight clarity, contrast, and noise level, thereby giving higher weights to high-quality features (features with higher confidence). In step A2214, the weighted fusion algorithm can employ linear or nonlinear weighting to weightedly fuse the clarity score, contrast score, and noise score with the corresponding weights obtained in step A2213 to obtain a final image quality score. This embodiment is equivalent to a method for adaptively determining weights based on local features of different regions in the expanded image. Therefore, this embodiment can further refine the image quality score obtained using the weighted fusion algorithm, thereby improving the accuracy of the image quality score and, in turn, enhancing the training effect and positioning accuracy of the joint cavity positioning model.

[0090] In some preferred embodiments, step A2211 includes:

[0091] A22111. For each expanded image, calculating a gradient magnitude image of the expanded image;

[0092] A22112, selecting the pixel with the largest gradient amplitude in the gradient amplitude image as the seed point;

[0093] A22113, expanding toward the neighborhood with the seed point as the center, and merging pixels in the neighborhood whose gradient magnitude difference with the seed point is less than a preset threshold into the current image block, until no pixel whose gradient magnitude difference with the seed point is less than the preset threshold exists in the gradient magnitude image;

[0094] A22114. Analyze whether there are undivided pixels in the gradient amplitude image. If so, select the pixel with the largest gradient amplitude among the undivided pixels as the new seed point and return to step A22113. If not, execute step 2212.

[0095] In step A22111, edge detection operators such as the Sobel operator, Prewitt operator, or Canny operator can be used to calculate the gradient magnitude image. These operators can effectively detect sudden changes in pixel grayscale values ​​in an image, thereby obtaining a gradient magnitude image. That is, step A22111 intentionally highlights the edge information of the image by calculating and expanding the gradient magnitude image. Since pixels with the largest gradient magnitude are typically located in regions rich in edge information, selecting such pixels as seed points ensures that the image block begins growing from the edge region of the image. Therefore, step A22112 selects the pixel with the largest gradient magnitude in the gradient magnitude image as the seed point. The expansion of the neighborhood from the seed point in step A22113 can be achieved by eight-neighborhood or four-neighborhood growth. The preset threshold value in this embodiment needs to be adjusted based on the actual application scenario and image characteristics to control the size and shape of the image block. It should be understood that this embodiment is equivalent to iteratively selecting the pixel with the largest gradient magnitude as the seed point and performing region growing with the seed point as the center to merge pixels with similar gradient magnitudes into the same image block. Since this embodiment can merge pixels with similar gradient amplitudes into the same image block, that is, the gradient features of the pixels in the same image block are similar, this embodiment can effectively improve the uniformity of the image features in the image block, thereby effectively further improving the accuracy of the image quality score.

[0096] In some preferred embodiments, step A22111 includes:

[0097] A221111. For each expanded image, convert the expanded image to the HSV color space and perform blurring on the V channel of the HSV color space to obtain a light intensity map.

[0098] A221112. Calculate the illumination compensation coefficient corresponding to each pixel in the expanded image according to the illumination intensity map;

[0099] A221113. Adjust the RGB value corresponding to each pixel in the expanded image according to the illumination compensation coefficient;

[0100] A221114. Calculate the gradient magnitude image of the expanded image.

[0101] Step A22111 is intended to preliminarily process the image illumination. Step A22111 separates the color information and brightness information of the image by converting the expanded image into the HSV color space, so that the brightness information can be processed separately. The V channel represents the brightness information, and this channel is blurred, for example, using Gaussian blur, in order to smooth the illumination intensity map and reduce noise interference, thereby obtaining the overall illumination intensity distribution of the image. Step A221112 is used to calculate the illumination compensation coefficient. Specifically, the illumination compensation coefficient of each pixel is calculated by the ratio of the average brightness value of the illumination intensity map to the brightness value of each pixel. That is, the calculation formula of the illumination compensation coefficient is: ; Where k(x,y) represents the illumination compensation coefficient of the pixel with coordinates (x,y) in the illumination intensity map, V mean represents the average brightness value of the illumination intensity map, and V(x,y) represents the brightness value of the pixel with coordinates (x,y) in the illumination intensity map. The illumination compensation coefficient of the pixel is negatively correlated with its brightness value. The higher the brightness value, the smaller the compensation coefficient, and the lower the brightness value, the larger the illumination compensation coefficient. It should be understood that the number of pixels contained in the illumination intensity map is equal to the number of pixels contained in the expanded image, that is, each pixel in the illumination intensity map corresponds to a pixel in the expanded image. Step A221113 is used to adjust the RGB value of the image according to the illumination compensation coefficient to correct the image brightness difference caused by uneven illumination. The adjusted RGB value is obtained by multiplying the original RGB value by the illumination compensation coefficient. The RGB value adjustment formula is:

[0102] ;

[0103] in, 、 and represents the RGB value of the pixel with coordinates (x, y) in the illumination intensity map after adjustment, and R(x, y), G(x, y) and B(x, y) represent the RGB value of the pixel with coordinates (x, y) in the illumination intensity map before adjustment. Step A221114 is used to calculate the gradient amplitude image of the expanded image after illumination compensation, providing a basis for subsequent image block division. Because before calculating the gradient amplitude image of the expanded image, this embodiment first obtains the illumination intensity map corresponding to the expanded image and adjusts the RGB values ​​of the pixels in the expanded image based on the illumination intensity map to perform illumination balancing on the expanded image, this embodiment can effectively reduce the impact of uneven illumination on the calculation of the gradient amplitude image, thereby effectively improving the accuracy of the gradient amplitude image.

[0104] In some preferred embodiments, step A25 includes:

[0105] A251. Obtaining a high-frequency sub-band energy intensity feature corresponding to each expanded image based on a wavelet transform, and obtaining a loss weight corresponding to each expanded image according to the high-frequency sub-band energy intensity feature and a first preset conversion relationship;

[0106] A252. Based on the adaptive regularization coefficient function, the regularization strength of the preliminary positioning model is dynamically adjusted according to the joint cavity morphology diversity index of the expanded image dataset. During the adjustment process, the Adam optimizer is used to minimize the cross entropy loss function of the preliminary positioning model by multiplying the cross entropy loss value of each expanded image by the corresponding loss weight to obtain the joint cavity positioning model.

[0107] The wavelet transform of step A251 can be a Haar wavelet transform, a Daubechies wavelet transform, or a Symlets wavelet transform. The wavelet transform of step A251 can decompose the expanded image into multiple sub-bands of different frequencies, and the high-frequency sub-band energy intensity features are extracted from these sub-bands. Since the noise in the image is mainly concentrated in the high-frequency part, the high-frequency sub-band energy intensity features obtained in step A251 can reflect the noise intensity and noise level of the expanded image. The first preset conversion relationship of this embodiment is preferably a pre-constructed mapping relationship between the high-frequency sub-band energy intensity features and the loss weight. Specifically, the high-frequency sub-band energy intensity features are negatively correlated with the loss weight, that is, this embodiment is equivalent to assigning a lower loss weight to an image with high noise intensity and a higher loss weight to an image with low noise intensity. Since in the process of regularization strength adjustment, this embodiment uses the Adam optimizer to minimize the cross entropy loss function of the preliminary positioning model by multiplying the cross entropy loss value of each expanded image by the corresponding loss weight, so that the model makes the model training pay more attention to low-noise images and less attention to the influence of high-noise images, therefore, this embodiment can effectively improve the robustness and positioning accuracy of the joint cavity positioning model.

[0108] In some specific embodiments, step A251 may further include: first, performing wavelet decomposition on each expanded image to obtain subband coefficients of multiple frequency bands; then, calculating the energy of each high-frequency subband, where the energy calculation method may be to calculate the sum of the squares of the subband coefficients; then, performing weighted summation on the energies of all high-frequency subbands to obtain a high-frequency subband energy intensity feature; finally, based on the high-frequency subband energy intensity feature, obtaining the loss weight corresponding to each expanded image through a preset conversion relationship, such as a linear function, a nonlinear function, or a lookup table. As a preferred embodiment, the first preset conversion relationship may be set to a Sigmoid function, the input of which is the normalized high-frequency subband energy intensity feature, and the output is a loss weight in the interval [0,1]. In this way, the higher the high-frequency subband energy intensity feature, the lower the loss weight.

[0109] In some preferred embodiments, step S2 includes:

[0110] S21, the feature points of the current frame image are matched with the feature points of the previous frame image to obtain a matching pair of feature points;

[0111] S22. Estimate the homography matrix of the current frame image relative to the previous frame image based on the matched feature point pairs using the RANSAC algorithm. The homography matrix describes the transformation relationship of the arthroscopic view:

[0112] S23, decomposing the homography matrix into a rotation matrix and a translation vector, and performing rotation and translation operations on the convolution kernel of the pre-trained joint cavity positioning model according to the rotation matrix and the translation vector, so that the joint cavity positioning model adapts to the current perspective;

[0113] S24. Input the enhanced image into the joint cavity positioning model to obtain joint cavity position information.

[0114] Step S21 can use SIFT, SURF or ORB feature matching algorithms to achieve feature point matching. These algorithms can extract distinctive feature points from the image, and establish the correspondence between the feature points of the current frame image and the previous frame image through descriptor matching to obtain matched feature point pairs. The RANSAC algorithm in step S2 is used to robustly estimate the homography matrix from the matched feature point pairs. The RANSAC algorithm effectively eliminates the interference of erroneous matching point pairs through random sampling and iterative optimization, thereby obtaining an accurate homography matrix. The homography matrix can describe the mapping relationship between the pixel points between the two frames of images and reflect the transformation of the arthroscopic viewing angle. Step S23 can use the singular value decomposition method to decompose the homography matrix to obtain a rotation matrix and a translation vector. The rotation matrix and the translation vector respectively represent the rotation and displacement changes of the arthroscopic viewing angle. The convolution kernel of the pre-trained joint cavity positioning model is a key component in the model for extracting image features. This embodiment can adapt the convolution kernel to the image features under the current perspective by applying a rotation matrix and a translation vector to the convolution kernel, thereby ensuring that the joint cavity positioning model can still effectively extract features when the perspective changes. This embodiment is equivalent to introducing a perspective adaptive adjustment mechanism in the joint cavity positioning process. Therefore, this embodiment can effectively solve the problem of reduced positioning accuracy due to changes in the arthroscopic perspective, thereby effectively improving the robustness and accuracy of the joint cavity positioning model.

[0115] In some specific embodiments, in step S21, the ORB feature matching algorithm is used to perform feature point matching to quickly and robustly find the feature point correspondence between the current frame image and the previous frame image. In step S22, the specific number of iterations of the RANSAC algorithm is set to 1000 times to ensure the accuracy and stability of the homography matrix estimation. In step S23, the homography matrix is ​​decomposed into a rotation matrix and a translation vector by a singular value decomposition method, and the decomposed rotation matrix and translation vector are directly applied to the convolution kernels of all convolution layers of the joint cavity positioning model. This embodiment can effectively realize the adaptive adjustment of the arthroscopic viewing angle to ensure that the joint cavity positioning model can still accurately locate the joint cavity when the arthroscopic viewing angle changes, thereby effectively improving the practical value of the joint cavity positioning model.

[0116] In some preferred embodiments, preprocessing further includes edge sharpening. Edge sharpening is a technique used to enhance edge contrast in an image, thereby making the boundaries between different regions clearer. This embodiment can sharpen the edges of the arthroscopic image to make the boundaries of the joint cavity clearer and less blurry, thereby enabling the joint cavity positioning model to more accurately identify and locate the joint cavity.

[0117] From the above, it can be seen that the joint cavity positioning method based on a deep learning model provided by the present application is equivalent to first training a pre-constructed deep learning model through a pre-collected general medical image data set so that the deep learning model can learn the general features of medical images and the basic knowledge of joint cavity positioning, and then dynamically adjust the regularization strength of the preliminary positioning model according to the joint cavity morphology diversity index of the expanded image data set based on an adaptive regularization coefficient function so that the deep learning model can learn the unique features of the joint cavity of rare bone and joint diseases while avoiding overfitting. Therefore, the joint cavity positioning model of the present application has good generalization ability and positioning accuracy, and the joint cavity positioning model can more accurately locate the joint cavity of rare bone and joint diseases, that is, the present application can effectively solve the problem that the joint cavity positioning model obtained based on the deep learning model cannot accurately locate the joint cavity of rare bone and joint diseases due to insufficient training data corresponding to rare bone and joint diseases, and the generalization ability and positioning accuracy of the joint cavity positioning model are reduced due to the diversity of joint cavity morphology.

[0118] Second, as Figure 2 As shown, the present application also provides a joint cavity positioning device based on a deep learning model, which is used to locate the joint cavity of rare bone and joint diseases. The joint cavity positioning device based on the deep learning model includes the following steps:

[0119] Enhanced image acquisition module 1, used for acquiring arthroscopic images and preprocessing the arthroscopic images to obtain enhanced images, wherein the preprocessing includes contrast enhancement;

[0120] Joint cavity position information acquisition module 2, which acquires joint cavity position information based on a pre-trained joint cavity positioning model and the enhanced image. The joint cavity positioning model is used to locate the joint cavity in the enhanced image and output the joint cavity position;

[0121] Pre-training of the joint cavity positioning model includes:

[0122] A1. Use a pre-collected general medical image dataset to train a pre-built deep learning model to obtain a preliminary positioning model.

[0123] A2. Expand the pre-collected arthroscopic image dataset of rare bone and joint diseases to obtain an expanded image dataset. Then, based on the adaptive regularization coefficient function, dynamically adjust the regularization strength of the preliminary positioning model according to the joint cavity morphology diversity index of the expanded image dataset to obtain a joint cavity positioning model.

[0124] A joint cavity positioning device based on a deep learning model provided in this application includes an enhanced image acquisition module 1 and a joint cavity position information acquisition module 2. The joint cavity positioning based on a deep learning model provided in this embodiment is used to execute a joint cavity positioning method based on a deep learning model provided in the first aspect above. The principle of the joint cavity positioning device based on a deep learning model provided in this embodiment is the same as the principle of the joint cavity positioning method based on a deep learning model provided in the first aspect above, and will not be discussed in detail here.

[0125] From the above, it can be seen that the joint cavity positioning method and device based on a deep learning model provided by the present application is equivalent to first training a pre-constructed deep learning model through a pre-collected general medical image data set so that the deep learning model learns the general features of medical images and the basic knowledge of joint cavity positioning, and then dynamically adjusts the regularization strength of the preliminary positioning model according to the joint cavity morphology diversity index of the expanded image data set based on an adaptive regularization coefficient function so that the deep learning model learns the unique features of the joint cavity of rare bone and joint diseases while avoiding overfitting. Therefore, the joint cavity positioning model of the present application has good generalization ability and positioning accuracy, and the joint cavity positioning model can more accurately locate the joint cavity of rare bone and joint diseases, that is, the present application can effectively solve the problem that the joint cavity positioning model obtained based on the deep learning model cannot accurately locate the joint cavity of rare bone and joint diseases due to insufficient training data corresponding to rare bone and joint diseases, and the generalization ability and positioning accuracy of the joint cavity positioning model are reduced due to the diversity of joint cavity morphology.

[0126] In the embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of the above-mentioned units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another robot, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some communication interface, the indirect coupling or communication connection of the device or unit can be electrical, mechanical or other forms.

[0127] In addition, the functional modules in each embodiment of the present application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0128] In this document, relational terms such as first and second, etc. are used merely to distinguish one entity or operation from another entity or operation, but do not necessarily require or imply any actual relationship or order between these entities or operations.

[0129] The above description is merely an embodiment of the present application and is not intended to limit the scope of protection of the present application. For those skilled in the art, various modifications and variations of the present application are possible. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application shall be included in the scope of protection of the present application.

Claims

1. A joint cavity positioning method based on a deep learning model, characterized in that: For locating the joint cavity of rare bone and joint diseases, the joint cavity positioning method based on the deep learning model includes the following steps: S1. Acquire an arthroscopic image and preprocess the arthroscopic image to obtain an enhanced image, wherein the preprocessing includes contrast enhancement; S2. Acquire joint cavity position information based on a pre-trained joint cavity positioning model and the enhanced image, wherein the joint cavity positioning model is used to locate the joint cavity in the enhanced image and output the joint cavity position; The pre-training of the joint cavity positioning model includes: A1. Use a pre-collected general medical image dataset to train a pre-built deep learning model to obtain a preliminary positioning model. A2. Expanding a pre-collected arthroscopic image dataset of rare bone and joint diseases to obtain an expanded image dataset, and then dynamically adjusting the regularization strength of the preliminary localization model based on an adaptive regularization coefficient function according to a joint cavity morphology diversity index of the expanded image dataset to obtain a joint cavity localization model; Step A2 includes: A21. Expanding a pre-collected rare bone and joint disease arthroscopic image dataset by randomly rotating, scaling, and flipping each image in the pre-collected rare bone and joint disease arthroscopic image dataset to obtain an expanded image dataset, wherein the expanded image dataset includes a plurality of expanded images; A22. Obtaining an image quality score for each of the expanded images using a pre-built image quality assessment model to obtain multiple image quality scores; A23. Normalizing all the image quality scores to the interval [0, 1] to obtain a quality weight corresponding to each of the expanded images, and obtaining a joint cavity morphological feature vector corresponding to each of the expanded images based on a local binary pattern histogram corresponding to the expanded image; A24. performing a weighted average calculation based on all the mass weights and all the joint cavity morphology feature vectors to obtain a joint cavity morphology diversity index; A25. Dynamically adjust the regularization strength of the preliminary positioning model according to the joint cavity morphology diversity index of the expanded image data set based on an adaptive regularization coefficient function to obtain a joint cavity positioning model.

2. The joint cavity positioning method based on a deep learning model according to claim 1, characterized in that: The image quality assessment model is used to extract the clarity features, contrast features, and noise level features of each of the expanded images, and output corresponding clarity scores, contrast scores, and noise scores. Step A22 includes: A221. For each of the expanded images, calculate the corresponding image quality score according to its corresponding clarity score, contrast score, and noise score based on a weighted fusion algorithm.

3. The joint cavity positioning method based on a deep learning model according to claim 2, characterized in that: Step A221 includes: A2211. For each of the expanded images, divide the expanded image into a plurality of non-overlapping image blocks; A2212. Obtaining a gradient distribution feature, a color histogram feature, and a texture feature of each image block, wherein the gradient distribution feature is obtained by calculating the gradient magnitude and direction of each image block, the color histogram feature is obtained by statistically analyzing the color distribution of each image block in the HSV color space, and the texture feature is obtained by calculating the gray-level co-occurrence matrix of each image block; A2213, calculating a clarity confidence, a contrast confidence, and a noise confidence of each image block based on the gradient distribution feature, the color histogram feature, and the texture feature, and normalizing the clarity confidence, the contrast confidence, and the noise confidence based on a Softmax function to obtain a clarity weight, a contrast weight, and a noise weight of a weighted fusion algorithm; A2214. Calculate the corresponding image quality score based on the corresponding clarity score, contrast score, and noise score based on the weighted fusion algorithm.

4. The joint cavity positioning method based on a deep learning model according to claim 3, characterized in that: Step A2211 includes: A22111. For each of the expanded images, calculate a gradient magnitude image of the expanded image; A22112, selecting a pixel with the largest gradient amplitude in the gradient amplitude image as a seed point; A22113, expanding toward a neighborhood with the seed point as the center, and merging pixels in the neighborhood whose gradient magnitude difference with the seed point is less than a preset threshold into the current image block, until no pixel in the gradient magnitude image has a gradient magnitude difference with the seed point less than the preset threshold; A22114. Analyze whether there are undivided pixels in the gradient amplitude image. If so, select the pixel with the largest gradient amplitude among the undivided pixels as the new seed point and return to step A22113. If not, execute step 2212.

5. The joint cavity positioning method based on a deep learning model according to claim 4, characterized in that: Step A22111 includes: A221111. For each expanded image, convert the expanded image to an HSV color space and perform blur processing on a V channel of the HSV color space to obtain a light intensity map. A221112. Calculate the illumination compensation coefficient corresponding to each pixel in the expanded image according to the illumination intensity map; A221113. Adjusting the RGB value corresponding to each pixel in the expanded image according to the illumination compensation coefficient; A221114. Calculate the gradient magnitude image of the expanded image.

6. The joint cavity positioning method based on a deep learning model according to claim 1, characterized in that: Step A25 includes: A251. Obtaining a high-frequency sub-band energy intensity feature corresponding to each of the expanded images based on a wavelet transform, and obtaining a loss weight corresponding to each of the expanded images according to the high-frequency sub-band energy intensity feature and a first preset conversion relationship; A252. Dynamically adjust the regularization strength of the preliminary positioning model according to the joint cavity morphology diversity index of the expanded image data set based on the adaptive regularization coefficient function, and use the Adam optimizer to minimize the cross entropy loss function of the preliminary positioning model during the adjustment process by multiplying the cross entropy loss value of each expanded image by the corresponding loss weight to obtain the joint cavity positioning model.

7. The joint cavity positioning method based on a deep learning model according to claim 1, characterized in that: Step S2 includes: S21, the feature points of the current frame image are matched with the feature points of the previous frame image to obtain a matching pair of feature points; S22. Estimate the homography matrix of the current frame image relative to the previous frame image based on the matched feature point pairs using the RANSAC algorithm. The homography matrix describes the transformation relationship of the arthroscopic viewing angle: S23, decomposing the homography matrix into a rotation matrix and a translation vector, and performing rotation and translation operations on the convolution kernel of the pre-trained joint cavity positioning model according to the rotation matrix and the translation vector, so that the joint cavity positioning model adapts to the current viewing angle; S24. Input the enhanced image into the joint cavity positioning model to obtain joint cavity position information.

8. The joint cavity positioning method based on a deep learning model according to claim 1, characterized in that: The pre-processing also includes edge sharpening.

9. A joint cavity positioning device based on a deep learning model, characterized in that: For locating the joint cavity of rare bone and joint diseases, the joint cavity positioning device based on the deep learning model includes the following steps: an enhanced image acquisition module, configured to acquire an arthroscopic image and preprocess the arthroscopic image to obtain an enhanced image, wherein the preprocessing includes contrast enhancement; a joint cavity position information acquisition module, which acquires joint cavity position information based on a pre-trained joint cavity positioning model and the enhanced image, wherein the joint cavity positioning model is used to locate the joint cavity in the enhanced image, and outputs the joint cavity position; The pre-training of the joint cavity positioning model includes: A1. Use a pre-collected general medical image dataset to train a pre-built deep learning model to obtain a preliminary positioning model. A2. Expanding a pre-collected arthroscopic image dataset of rare bone and joint diseases to obtain an expanded image dataset, and then dynamically adjusting the regularization strength of the preliminary localization model based on an adaptive regularization coefficient function according to a joint cavity morphology diversity index of the expanded image dataset to obtain a joint cavity localization model; Step A2 includes: A21. Expanding a pre-collected rare bone and joint disease arthroscopic image dataset by randomly rotating, scaling, and flipping each image in the pre-collected rare bone and joint disease arthroscopic image dataset to obtain an expanded image dataset, wherein the expanded image dataset includes a plurality of expanded images; A22. Obtaining an image quality score for each of the expanded images using a pre-built image quality assessment model to obtain multiple image quality scores; A23. Normalizing all the image quality scores to the interval [0, 1] to obtain a quality weight corresponding to each of the expanded images, and obtaining a joint cavity morphological feature vector corresponding to each of the expanded images based on a local binary pattern histogram corresponding to the expanded image; A24. performing a weighted average calculation based on all the mass weights and all the joint cavity morphology feature vectors to obtain a joint cavity morphology diversity index; A25. Dynamically adjust the regularization strength of the preliminary positioning model according to the joint cavity morphology diversity index of the expanded image data set based on an adaptive regularization coefficient function to obtain a joint cavity positioning model.

Images