Knee joint skeleton segmentation method and device based on deep learning network HuvNet
Through the deep learning network HuvNet combined with representation learning and specific preprocessing technology, the problem of poor knee bone segmentation in MRI images is solved, achieving more efficient and accurate knee bone segmentation, which is suitable for medical imaging analysis.
Patent Information
- Application Number
- CN202411408995.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-10
- Publication Date
- 2025-01-10
AI Technical Summary
Existing deep learning methods segment knee bones, especially patella and cartilage, in MRI images, are ineffective and have low segmentation efficiency.
Using the knee joint bone segmentation method based on the deep learning network HuvNet, a simple and efficient encoder and decoder are constructed through the combination of convolutional neural networks, and MRI data training is performed to automatically segment the femur, tibia, patella and cartilage, and truncate using pixel intensity truncation. The normalization process removes noise, introduces the residual module to alleviate the gradient vanishing problem, and uses RECON_LOSS and SEGM_LOSS loss function optimization models.
It improves the accuracy and efficiency of knee bone segmentation, enhances the expression ability and generalization ability of the model, ensures the accuracy and stability of the segmentation results, and is suitable for complex medical imaging analysis tasks.
Smart Images
Figure CN119295482A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of medical image processing, and in particular to a knee joint bone segmentation method and device based on a deep learning network HuvNet. Background Art
[0002] Currently, image segmentation technology can accurately segment structures such as femur, tibia, patella and cartilage from medical images (such as MRI), which can help doctors more accurately diagnose related diseases such as fractures, arthritis, cartilage damage, etc. For example, detailed morphological data can be extracted from them, such as bone size, shape, position relationship, and cartilage thickness and integrity, which helps doctors assess the severity of the disease and formulate more reasonable treatment plans.
[0003] There are many methods for image segmentation of knee bones, such as manual segmentation, semi-automatic segmentation, and fully automatic segmentation. Deep learning methods are also gradually being used in the field of medical image segmentation, such as using UNET and deeplab neural network structures. However, most of these methods use data sets for femur and tibia segmentation on CT data. As for patella and cartilage, they are limited by the data type and cannot be segmented well. Some deep learning segmentation methods using MRI use network structures that cannot learn the characteristics of the data well, and the segmentation effect has a large room for optimization. In summary, the segmentation efficiency of the current segmentation network is low. Summary of the invention
[0004] The purpose of the present invention is to provide a knee joint bone segmentation method and device based on a deep learning network HuvNet. Through a new deep learning neural network structure HuvNet, a convolutional neural network is used, the representation learning process is combined, and a simple and efficient encoder and decoder are constructed. Data training is performed on MRI data to automatically segment the femur, tibia, patella and cartilage.
[0005] The present invention provides a knee joint bone segmentation method based on a deep learning network HuvNet, comprising:
[0006] Acquire MRI image data containing knee bones of different patients;
[0007] Preprocessing the collected MRI image data and dividing them into a preprocessed training set and a test set according to a preset ratio;
[0008] Construct a bone segmentation prediction model based on the deep learning network HuvNet combined with the representation learning process;
[0009] The preprocessed training set is input into the bone segmentation prediction model for training to obtain a trained neural network model;
[0010] The test set is input into the trained bone segmentation prediction model to predict the corresponding bone segmentation results on the knee joint.
[0011] Preferably, the preprocessing of the collected MRI image data includes:
[0012] The pixel intensity value hu of the original MRI image data is truncated to remove noise and information irrelevant to the knee joint bones;
[0013] The truncated MRI image data is normalized so that different features are on the same scale. The normalization formula is as follows:
[0014]
[0015] Among them, hu_min is the minimum pixel intensity, hu_max is the maximum pixel intensity, and the normalized value range is [-1,1].
[0016] Preferably, the preprocessing of the collected MRI image data further comprises:
[0017] Cropping and adjusting the original MRI image data;
[0018] Each MRI image is cropped to a preset pixel size to fit the input layer of the model;
[0019] The number of slices of each MRI image sequence is configured as a preset number. If the actual number of slices is less than the preset number, it is padded with 0; if the actual number of slices exceeds the preset number, only the preset number of slices in the middle is retained to ensure that each sample has the same dimension.
[0020] Preferably, the process of constructing a skeleton segmentation prediction model based on a deep learning network HuvNet combined with representation learning includes:
[0021] A neural network model HuvNet is configured to include a plurality of encoders and a plurality of decoders and to combine representation learning, wherein the neural network model HuvNet at least includes a first encoder, a first decoder, a second encoder, and a second decoder connected in sequence;
[0022] The original MRI image data is input into the neural network, and a series of target features are obtained after being encoded by the first encoder;
[0023] From the Gaussian distribution After sampling, the target features are reparameterized for representation learning, and the representation formula is:
[0024] z=μ+σ⊙∈;
[0025] where z is the latent variable sampled from the latent space, μ is the mean of the first encoder output, σ is the variance of the first encoder output, and ∈ is a random variable sampled from a standard normal distribution. The noise sampled in .
[0026] Preferably, the process of constructing a bone segmentation prediction model based on the deep learning network HuvNet combined with representation learning also includes:
[0027] The target features after representation learning are input into the first decoder for decoding, and the loss of the decoded image and the annotated image is calculated to obtain the RECON_LOSS reconstruction loss result to measure the difference between the image output by the first decoder and the annotated image;
[0028] The target features output by the first decoder are further passed through the second encoder and the second decoder. The decoding result is compared with the annotated mask image for loss calculation, and the SEGM_LOSS segmentation loss result is obtained to measure the difference between the segmentation result output by the network and the annotated true segmentation result.
[0029] At the same time, gradient descent is performed on the RECON_LOSS reconstruction loss and SEGM_LOSS segmentation loss to adjust and optimize the model parameters to obtain the final optimized bone segmentation prediction model.
[0030] Preferably, the encoder includes 4 OPS modules, wherein each OPS module includes a CONV_BLOCK1 residual module, a CONV_BLOCK2 residual module and a downsampling module, and the downsampling module is implemented using convolution, and the stride is set to 2;
[0031] The decoder includes 4 OPS modules, wherein each OPS module includes a CONV_BLOCK1 residual module, a CONV_BLOCK2 residual module and an upsampling module, and the upsampling module is implemented using nearest neighbor interpolation.
[0032] As a preference,
[0033] The RECON_LOSS reconstruction loss includes reconstruction loss and KL divergence, where:
[0034] The reconstruction loss implementation formula is:
[0035]
[0036] Among them, x i is the ith original input sample, is the i-th sample generated, and N is the total number of samples;
[0037] The KL divergence loss implementation formula is:
[0038]
[0039] Among them, μ j and σ j are the mean and standard deviation of the encoder output, respectively, and N is the total number of samples;
[0040] The segmentation loss used by the SEGM_LOSS segmentation loss is dice loss, and the implementation formula is:
[0041]
[0042] Among them, C represents the number of samples, L Dice Represents the loss function, which is used to evaluate the performance of the model on the image segmentation task, Xc represents the true label, Yc represents the predicted result, and ∈ is a constant to prevent meaningless division.
[0043] The present invention also provides a knee joint bone segmentation device based on a deep learning network HuvNet, and implements a knee joint bone segmentation method based on a deep learning network HuvNet as described in an embodiment of the present invention, including:
[0044] An acquisition module, used for acquiring MRI image data containing knee joint bones of different patients;
[0045] A preprocessing module is used to preprocess the collected MRI image data and divide it into a preprocessed training set and a test set according to a preset ratio;
[0046] Model building module, used to build a skeleton segmentation prediction model based on the deep learning network HuvNet combined with the representation learning process;
[0047] A training module is used to input the preprocessed training set into the bone segmentation prediction model for training to obtain a trained neural network model;
[0048] The prediction module is used to input the test set into the trained bone segmentation prediction model to predict the corresponding bone segmentation results on the knee joint.
[0049] The present invention also provides an electronic device, comprising:
[0050] A memory, the memory being used to store a processing program;
[0051] A processor, when executing the processing program, implements the knee joint bone segmentation method based on the deep learning network HuvNet as described in the present invention.
[0052] The present invention also provides a readable storage medium, on which a processing program is stored. When the processing program is executed by a processor, the knee joint bone segmentation method based on the deep learning network HuvNet as described in the present invention is implemented.
[0053] With respect to the prior art, the present invention has the following beneficial effects:
[0054] The present invention effectively removes noise and irrelevant information by truncating and normalizing the pixel intensity values of MRI image data, making different features on the same scale, which helps to improve the stability and accuracy of subsequent model training. Each MRI image is cropped to a preset pixel size, and the number of slices in the sequence is configured to a preset number to ensure that each sample has the same dimension, which not only simplifies the processing of model input, but also reduces the amount of calculation and improves training efficiency.
[0055] The present invention combines the deep learning network HuvNet with the representation learning process, which can effectively extract image features and perform segmentation prediction, so that the model has stronger expression ability when processing complex image tasks; the introduction of residual modules in the encoder and decoder helps to alleviate the gradient vanishing problem in the deep network and improve the training effect of the model; the downsampling module in the encoder is implemented using convolution, and the upsampling module in the decoder is implemented using nearest neighbor interpolation, which helps to maintain the spatial information of the image and improve the accuracy of the segmentation result.
[0056] The RECON_LOSS reconstruction loss adopted in the present invention includes reconstruction loss and KL divergence, which is used to measure the difference between the image output by the model and the annotated image, which helps to optimize the reconstruction ability of the model and make the prediction result closer to the annotated image; the SEGM_LOSS segmentation loss adopted, that is, Dice Loss is used as the segmentation loss, which is used to evaluate the performance of the model in the image segmentation task, taking into account the overlapping area between the predicted label and the true label, and can more accurately reflect the segmentation effect of the model.
[0057] The present invention can effectively improve the performance of the model and make it better adapt to the task of knee bone segmentation by performing gradient descent on RECON_LOSS reconstruction loss and SEGM_LOSS segmentation loss at the same time, adjusting and optimizing model parameters; in the process of constructing the bone segmentation prediction model, combined with the process of representation learning, by sampling from Gaussian distribution and re-parameterizing the target features, effective learning of latent variables is achieved, which helps to improve the generalization ability and robustness of the model.
[0058] In summary, the knee joint bone segmentation method based on the deep learning network HuvNet of the present invention performs well in data preprocessing, network structure design, loss function selection, model training and optimization, etc., and can effectively improve the accuracy and efficiency of knee joint bone segmentation, providing strong technical support for the field of medical image analysis. BRIEF DESCRIPTION OF THE DRAWINGS
[0059] Figure 1-2 A flowchart of a knee joint bone segmentation method based on a deep learning network HuvNet provided by the present invention;
[0060] Figure 3 This is a schematic diagram of the segmentation result in an embodiment of the present invention;
[0061] Figure 4 An example diagram of a neural network model HuvNet combined with representation learning in an embodiment of the present invention;
[0062] Figure 5-6 This is an example diagram of the conv_block structure of the encoder and decoder corresponding to the model in the embodiment of the present invention;
[0063] Figure 7 This is a diagram showing a specific structure of an encoder according to an embodiment of the present invention;
[0064] Figure 8 2 is a diagram showing a specific structure of a decoder in an embodiment of the present invention. DETAILED DESCRIPTION
[0065] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0066] Embodiment 1
[0067] like Figure 1-2 As shown, the present invention provides a knee joint bone segmentation method based on a deep learning network HuvNet, comprising:
[0068] S1: Collect MRI image data containing the knee bones of different patients;
[0069] S2: preprocessing the acquired MRI image data, and dividing it into a preprocessed training set and a test set according to a preset ratio;
[0070] In step S2, preprocessing the collected MRI image data includes:
[0071] The pixel intensity value hu of the original MRI image data is truncated to remove noise and information irrelevant to the knee joint bones;
[0072] The truncated MRI image data is normalized so that different features are on the same scale. The normalization formula is as follows:
[0073]
[0074] Among them, hu_min is the minimum pixel intensity, hu_max is the maximum pixel intensity, and the normalized value range is [-1,1].
[0075] Since the collected original data is large, in order to adapt the following network model, the preprocessing of the collected MRI image data in step S2 further includes:
[0076] Cropping and adjusting the original MRI image data;
[0077] Each MRI image is cropped to a preset pixel size to fit the input layer of the model;
[0078] The number of slices of each MRI image sequence is configured to a preset number. If the actual number of slices is less than the preset number, it is padded with 0; if the actual number of slices exceeds the preset number, only the preset number of slices in the middle is retained to ensure that each sample has the same dimension. Therefore, the location of the bone area is cropped to a size of 256*256*, and the number of slices is set to 32. If it is insufficient, it is padded with 0, and if it exceeds 32, the middle part. This article uses 12GB of GPU memory. After testing, this data size is the maximum data that this article can process under the neural network structure. After that, the data is sent to the neural network proposed in this article for training, and the trained model is used for inference. The dice coefficient reaches 0.925. The results are as follows Figure 3 shown.
[0079] S3: Construct a skeleton segmentation prediction model based on the deep learning network HuvNet combined with the representation learning process;
[0080] See also Figure 4 As shown, the specific implementation steps of the new neural network architecture HuvNet for skeleton segmentation in this embodiment are as follows:
[0081] A neural network model HuvNet is configured to include a plurality of encoders and a plurality of decoders and to combine representation learning, wherein the neural network model HuvNet at least includes a first encoder encoder1, a first decoder decoder1, a second encoder encoder3, and a second decoder decoder3 connected in sequence;
[0082] The original MRI image data is input into the neural network, and after being encoded by the first encoder encoder1, a series of target features are obtained. The target features include morphological data, such as bone size, shape, position relationship, and cartilage thickness and integrity;
[0083] From the Gaussian distribution After sampling, the target features are reparameterized for representation learning, and the representation formula is:
[0084] z=μ+σ⊙∈;
[0085] Where z is the latent variable sampled from the latent space, μ is the mean of the output of the first encoder encoder1, σ is the variance of the output of the first encoder encoder1, and ∈ is the variance of the output from the standard normal distribution. The noise sampled in .
[0086] The target features after representation learning are input into the first decoder decoder1 for decoding, and the loss of the decoded image and the annotated image is calculated to obtain the RECON_LOSS reconstruction loss result to measure the difference between the image output by the first decoder decoder1 and the annotated image; the annotated image raw image is the original image excluding the effective area of the annotation mask. The target features output by the first decoder decoder1 are further passed through the second encoder encoder3 and the second decoder decoder3, and the decoding result and the annotated mask image are calculated to obtain the SEGM_LOSS segmentation loss result to measure the difference between the segmentation result output by the network and the annotated true segmentation result;
[0087] At the same time, gradient descent is performed on the RECON_LOSS reconstruction loss and SEGM_LOSS segmentation loss to adjust and optimize the model parameters to obtain the final optimized bone segmentation prediction model.
[0088] At the same time, in order to combine the actual situation, this embodiment also provides optional branches, such as Figure 4 As shown in the middle shadow, the original input is directly used for representation learning to reconstruct the original image and help extract data features. This can further improve segmentation accuracy at the expense of training time.
[0089] like Figure 5-6As shown, the encoder and decoder used in this embodiment are similar, but the convolution module CONV_BLOCK used is a residual module, where Dwise represents a depth-separable convolution, which can significantly reduce the number of parameters and improve the training and reasoning speed. The normalization method used is group_norm, because in the test, this paper found that when batch_size is small, group_norm works better than batch_norm.
[0090] like Figure 7 As shown, the encoder includes 4 OPS modules, wherein each OPS module includes a CONV_BLOCK1 residual module, a CONV_BLOCK2 residual module and a downsampling module, and the downsampling module is implemented using convolution, and the stride is set to 2;
[0091] like Figure 8 As shown, the decoder includes 4 OPS modules, wherein each OPS module includes a CONV_BLOCK1 residual module, a CONV_BLOCK2 residual module and an upsampling module, and the upsampling module is implemented using nearest neighbor interpolation.
[0092] In this embodiment, the use of the loss function is described in detail as follows:
[0093] The RECON_LOSS reconstruction loss includes reconstruction loss and KL divergence, where:
[0094] The reconstruction loss implementation formula is:
[0095]
[0096] Among them, x i is the ith original input sample, is the i-th sample generated, and N is the total number of samples;
[0097] The KL divergence loss implementation formula is:
[0098]
[0099] Among them, μ j and σ j are the mean and standard deviation of the encoder output, respectively, and N is the total number of samples;
[0100] The segmentation loss used by the SEGM_LOSS segmentation loss is dice loss, and the implementation formula is:
[0101]
[0102] Among them, C represents the number of samples, LDice Represents the loss function, which is used to evaluate the performance of the model in the image segmentation task. Xc represents the true label, Yc represents the predicted result, and ∈ is a very small constant set to prevent meaningless division.
[0103] In summary, the neural network structure proposed in this embodiment processes and analyzes MRI images through encoder and decoder, uses optional branches to improve segmentation accuracy, and adopts a specific loss function to optimize model performance.
[0104] S4: The preprocessed training set is input into the bone segmentation prediction model for training to obtain a trained neural network model. During the neural network training process, the model parameters are optimized by performing gradient descent on the reconstruction loss and segmentation loss at the same time, and finally a neural network model that can accurately reconstruct the input image and accurately segment the image is obtained. During the training process, two loss functions, RECON_LOSS and SEGM_LOSS, need to be calculated. RECON_LOSS reconstruction loss includes reconstruction loss and KL divergence, and SEGM_LOSS segmentation loss is dice loss. Through optimization algorithms such as gradient descent, the model parameters are continuously adjusted to minimize the loss function, thereby obtaining the optimal segmentation model. The trained model is tested and evaluated using the test data set, and indicators such as the dice coefficient are calculated to evaluate the performance of the model. If the performance meets the expected requirements, the model can be applied to the actual knee bone segmentation task.
[0105] S5: Input the test set into the trained bone segmentation prediction model to predict the corresponding bone segmentation results on the knee joint, such as accurately segmenting the femur, tibia, patella and cartilage from MRI medical images through image segmentation technology.
[0106] Embodiment 2
[0107] Based on the same concept, the present invention also provides a knee joint bone segmentation device based on the deep learning network HuvNet, which implements the knee joint bone segmentation method based on the deep learning network HuvNet as described in the first embodiment above, including:
[0108] An acquisition module, used for acquiring MRI image data containing knee joint bones of different patients;
[0109] A preprocessing module is used to preprocess the collected MRI image data and divide it into a preprocessed training set and a test set according to a preset ratio;
[0110] Model building module, used to build a skeleton segmentation prediction model based on the deep learning network HuvNet combined with the representation learning process;
[0111] A training module is used to input the preprocessed training set into the bone segmentation prediction model for training to obtain a trained neural network model;
[0112] The prediction module is used to input the test set into the trained bone segmentation prediction model to predict the corresponding bone segmentation results on the knee joint.
[0113] The specific principles, contents and implementation methods of the above modules are as described in Example 1 and will not be repeated here.
[0114] Finally, in order to apply the above-mentioned knee joint bone segmentation method based on the deep learning network HuvNet to a length quantification system, device or equipment with relevant hardware conditions, the present application also provides a computer-readable storage medium, in which a computer program is loaded, and when the computer program is executed by a computer, it realizes the functions of the corresponding method embodiment as described above.
[0115] Similarly, the present application also protects a computer-readable storage medium loaded with a computer program for implementing the knee joint bone segmentation method based on the deep learning network HuvNet.
[0116] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program. The computer program includes one or more computer programs. When the computer program is loaded and executed on a computer, the process or function described in the embodiment of the present disclosure is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer program can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer program can be transmitted from a website site, computer, server or data center by wired (e.g., coaxial cable, optical fiber, DDL (Digital DubDriberLine, digital subscriber line)) or wireless (e.g., infrared, wireless, microwave, etc.) mode to another website site, computer, server or data center. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more available media integrations. The available medium may be a magnetic medium (eg, a floppy disk, a hard disk, a magnetic tape), an optical medium (eg, a high-density DVD (Digital Video DiDD)), or a semiconductor medium (eg, a solid-state drive (SSD)).
[0117] It should be noted that, in the above-mentioned embodiments, the terms used herein are only for describing specific exemplary embodiments, and are not intended to be restrictive. As used herein, the singular forms "one", "an" and "the or described" may be intended to also include plural forms, unless the context clearly indicates otherwise. The terms "include", "comprise" and "have" are inclusive, and therefore specify the existence of stated features, wholes, steps, operations, elements and / or parts, but do not exclude the existence or addition of one or more other features, wholes, steps, operations, elements, parts and / or their groups. The method steps, processes and operations described herein should not be interpreted as necessarily requiring the method steps, processes and operations to be performed in the specific order discussed or shown, unless it is specifically indicated that the steps are to be performed in the set order. It should also be understood that additional or alternative steps may be adopted.
[0118] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A knee joint bone segmentation method based on deep learning network HuvNet, characterized in that: include: Acquire MRI image data containing knee bones of different patients; Preprocessing the collected MRI image data and dividing them into a preprocessed training set and a test set according to a preset ratio; Construct a bone segmentation prediction model based on the deep learning network HuvNet combined with the representation learning process; The preprocessed training set is input into the bone segmentation prediction model for training to obtain a trained neural network model; The test set is input into the trained bone segmentation prediction model to predict the corresponding bone segmentation results on the knee joint.
2. The knee joint bone segmentation method based on the deep learning network HuvNet according to claim 1 is characterized in that: The preprocessing of the collected MRI image data includes: The pixel intensity value hu of the original MRI image data is truncated to remove noise and information irrelevant to the knee joint bones; The truncated MRI image data is normalized so that different features are on the same scale. The normalization formula is as follows: Among them, hu_min is the minimum pixel intensity, hu_max is the maximum pixel intensity, and the normalized value range is [-1,1].
3. The knee joint bone segmentation method based on the deep learning network HuvNet according to claim 1 is characterized in that: The preprocessing of the collected MRI image data further includes: Cropping and adjusting the original MRI image data; Each MRI image is cropped to a preset pixel size to fit the input layer of the model; The number of slices of each MRI image sequence is configured as a preset number. If the actual number of slices is less than the preset number, it is padded with 0; if the actual number of slices exceeds the preset number, only the preset number of slices in the middle are retained to ensure that each sample has the same dimension.
4. The knee joint bone segmentation method based on the deep learning network HuvNet according to claim 1 is characterized in that: The process of building a bone segmentation prediction model based on the deep learning network HuvNet combined with representation learning includes: A neural network model HuvNet is configured to include a plurality of encoders and a plurality of decoders and to combine representation learning, wherein the neural network model HuvNet at least includes a first encoder, a first decoder, a second encoder, and a second decoder connected in sequence; The original MRI image data is input into the neural network, and a series of target features are obtained after being encoded by the first encoder; From the Gaussian distribution After sampling, the target features are reparameterized for representation learning, and the representation formula is: z=μ+σ⊙∈; where z is the latent variable sampled from the latent space, μ is the mean of the first encoder output, σ is the variance of the first encoder output, and ∈ is a random variable sampled from a standard normal distribution. The noise sampled in .
5. The knee joint bone segmentation method based on the deep learning network HuvNet according to claim 4 is characterized in that: The process of building a bone segmentation prediction model based on the deep learning network HuvNet combined with representation learning also includes: The target features after representation learning are input into the first decoder for decoding, and the loss of the decoded image and the annotated image is calculated to obtain the RECON_LOSS reconstruction loss result to measure the difference between the image output by the first decoder and the annotated image; The target features output by the first decoder are further passed through the second encoder and the second decoder. The decoding result is compared with the annotated mask image for loss calculation, and the SEGM_LOSS segmentation loss result is obtained to measure the difference between the segmentation result output by the network and the annotated true segmentation result. At the same time, gradient descent is performed on the RECON_LOSS reconstruction loss and SEGM_LOSS segmentation loss to adjust and optimize the model parameters to obtain the final optimized bone segmentation prediction model.
6. The knee joint bone segmentation method based on the deep learning network HuvNet according to claim 4 is characterized in that: The encoder includes 4 OPS modules, wherein each OPS module includes a CONV_BLOCK1 residual module, a CONV_BLOCK2 residual module and a downsampling module, and the downsampling module is implemented using convolution, and the stride is set to 2; The decoder includes 4 OPS modules, wherein each OPS module includes a CONV_BLOCK1 residual module, a CONV_BLOCK2 residual module and an upsampling module, and the upsampling module is implemented using nearest neighbor interpolation.
7. The knee joint bone segmentation method based on the deep learning network HuvNet according to claim 5 is characterized in that: The RECON_LOSS reconstruction loss includes reconstruction loss and KL divergence, where: The reconstruction loss implementation formula is: Among them, x i is the ith original input sample, is the i-th sample generated, and N is the total number of samples; The KL divergence loss implementation formula is: Among them, μ j and σ j are the mean and standard deviation of the encoder output, respectively, and N is the total number of samples; The segmentation loss used by the SEGM_LOSS segmentation loss is dice loss, and the implementation formula is: Among them, C represents the number of samples, L Dice Represents the loss function, which is used to evaluate the performance of the model on the image segmentation task, Xc represents the true label, Yc represents the predicted result, and ∈ is a constant to prevent meaningless division.
8. A knee joint bone segmentation device based on the deep learning network HuvNet, characterized in that: Implementing the knee joint bone segmentation method based on the deep learning network HuvNet as described in any one of claims 1 to 7, comprising: An acquisition module, used for acquiring MRI image data including knee joint bones of different patients; A preprocessing module is used to preprocess the collected MRI image data and divide it into a preprocessed training set and a test set according to a preset ratio; Model building module, used to build a skeleton segmentation prediction model based on the deep learning network HuvNet combined with the representation learning process; A training module is used to input the preprocessed training set into the bone segmentation prediction model for training to obtain a trained neural network model; The prediction module is used to input the test set into the trained bone segmentation prediction model to predict the corresponding bone segmentation results on the knee joint.
9. An electronic device, characterized in that: include: A memory, the memory being used to store a processing program; A processor, wherein when executing the processing program, the processor implements the knee joint bone segmentation method based on the deep learning network HuvNet as described in any one of claims 1 to 7.
10. A readable storage medium, characterized in that: The readable storage medium stores a processing program, and when the processing program is executed by the processor, the knee joint bone segmentation method based on the deep learning network HuvNet as described in any one of claims 1 to 7 is implemented.