Deep learning fat muscle segmentation method and system based on Slicer

Abstract

The invention provides a Slicer-based deep learning fat muscle segmentation method and system, and is applied to the technical field of data processing. The method comprises the following steps: preprocessing preoperative CT image data of a target patient to generate skeletal muscle image features, subcutaneous adipose tissue image features and visceral adipose tissue image features; preprocessing the training sample set, adjusting a window width and a window level, performing normalization processing, and generating target training image data; training a preset fat muscle segmentation model based on a deep learning network based on the target training image data to generate a target fat muscle segmentation model; processing the skeletal muscle image features, the subcutaneous adipose tissue image features and the visceral adipose tissue image features based on a target adipose muscle segmentation model to generate target image segmentation information; and processing the target image segmentation information and the gene detection information of the target patient to generate preoperative early warning information of the target patient.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and particularly relates to a deep learning fat and muscle segmentation method and system based on Slicer. Background Art

[0002] In the medical field, accurate preoperative assessment of a patient's fat and muscle conditions is extremely crucial for surgical plan formulation and prognosis judgment. Currently, there are many deficiencies in processing preoperative relevant data of patients and conducting fat and muscle analysis. From the perspective of image data processing, the original CT images are vulnerable to noise interference, resulting in poor image quality, which increases the difficulty of subsequent observation and analysis of fat and muscle tissues. For example, noise may blur the boundaries between muscles and fat, affecting doctors' accurate judgment of tissue morphology and scope. Moreover, there is no unified standard for the image data obtained by different devices and parameters. When performing operations such as window width and window level adjustment, there are no mature and universal specifications, resulting in large differences in processing results and making it difficult to ensure the accuracy and consistency of analysis.

[0003] In terms of fat and muscle segmentation models, existing segmentation models have insufficient segmentation accuracy when facing complex human fat and muscle structures. For example, some models cannot accurately distinguish the boundaries between subcutaneous fat and visceral fat, and there are also defects in the detailed segmentation of skeletal muscles, unable to meet the clinical requirements for accurate segmentation. Moreover, the data samples used for model training often have problems such as single features and insufficient quantity, making the trained models have poor generalization ability and difficult to adapt to the individual differences of different patients.

[0004] It should be noted that the information disclosed in the above background art section is only used to enhance the understanding of the background of the present disclosure, and thus may include information that does not constitute the prior art known to those of ordinary skill in the art. Summary of the Invention

[0005] The purpose of this application is to provide a deep learning fat and muscle segmentation method and system based on Slicer, which can at least overcome the problems existing in the prior art to a certain extent. By acquiring data such as preoperative CT images and gene detection information of patients, and performing preprocessing such as denoising, window width and window level adjustment, and normalization, tissue image features and target training image data are generated. The trained model is used to process the image features to obtain target image segmentation information. Combining gene detection information, multi-faceted analysis from aspects such as tissue morphology and image quantification is carried out to generate risk level information and preoperative warnings, assisting doctors in comprehensively understanding the patient's condition, aiming to achieve accurate preoperative assessment.

[0006] Other features and advantages of this application will become apparent through the following detailed description, or will be partially learned through the practice of the present invention.

[0007] According to one aspect of the present application, a deep learning fat and muscle segmentation method based on Slicer is provided, including: obtaining preoperative CT image data of a target patient, genetic testing information of the target patient, a preset fat and muscle segmentation model based on a deep learning network, and a training sample set; preprocessing the preoperative CT image data of the target patient to generate skeletal muscle image features, subcutaneous fat tissue image features, and visceral fat tissue image features; preprocessing the training sample set, adjusting the window width and window level and performing normalization processing to generate target training image data; training the preset fat and muscle segmentation model based on the deep learning network based on the target training image data to generate a target fat and muscle segmentation model; processing the skeletal muscle image features, subcutaneous fat tissue image features, and visceral fat tissue image features based on the target fat and muscle segmentation model to generate target image segmentation information, where the target image segmentation information includes the segmentation results of the skeletal muscle image, subcutaneous fat tissue image, and visceral fat tissue image respectively, and the regional surface sum and average density; processing the target image segmentation information and the genetic testing information of the target patient to generate preoperative warning information of the target patient.

[0008] According to another aspect of the present application, a deep learning fat and muscle segmentation device based on Slicer, characterized by including: an acquisition module, configured to obtain preoperative CT image data of a target patient, genetic testing information of the target patient, a preset fat and muscle segmentation model based on a deep learning network, and a training sample set; a processing module, configured to preprocess the preoperative CT image data of the target patient to generate skeletal muscle image features, subcutaneous fat tissue image features, and visceral fat tissue image features; preprocess the training sample set, adjust the window width and window level and perform normalization processing to generate target training image data; train the preset fat and muscle segmentation model based on the deep learning network based on the target training image data to generate a target fat and muscle segmentation model; process the skeletal muscle image features, subcutaneous fat tissue image features, and visceral fat tissue image features based on the target fat and muscle segmentation model to generate target image segmentation information; process the target image segmentation information and the genetic testing information of the target patient to generate preoperative warning information of the target patient.

[0009] According to still another aspect of the present application, an electronic device, characterized by including: a first processor; and a memory for storing executable instructions of the first processor; wherein, the first processor is configured to execute the above-mentioned deep learning fat and muscle segmentation method based on Slicer by executing the executable instructions.

[0010] According to another aspect of the present application, there is provided a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a second processor, the above-mentioned Slicer-based deep learning fat and muscle segmentation method is implemented.

[0011] The Slicer-based deep learning fat and muscle segmentation method and system provided by the present application obtain data such as preoperative CT images and gene detection information of patients by a server, and generate tissue image features and target training image data through preprocessing such as denoising, window width and window level adjustment, and normalization. The trained model is used to process the image features to obtain target image segmentation information. Finally, combined with gene detection information, multi-faceted analysis such as tissue morphology and image quantification is performed to generate risk level information and preoperative warning, assisting doctors to comprehensively understand the patient's condition, formulate a more reasonable surgical plan, and improve the accuracy and effectiveness of medical decisions, aiming to achieve accurate preoperative assessment.

[0012] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and cannot limit the present disclosure. Brief Description of the Drawings

[0013] Figure 1 A flowchart showing a Slicer-based deep learning fat and muscle segmentation method provided by an embodiment of the present application; Figure 2 A schematic structural diagram showing a target fat and muscle segmentation model provided by an embodiment of the present application; Figure 3 A schematic structural diagram showing a Slicer-based deep learning fat and muscle segmentation device provided by an embodiment of the present application. Detailed Embodiment

[0014] The following describes the preferred embodiments of the present invention with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are only for explaining and illustrating the present invention, and are not used to limit the present invention.

[0015] The following combines Figure 1 to describe a Slicer-based deep learning fat and muscle segmentation method according to an exemplary embodiment of the present application. In one embodiment, the present application also proposes a Slicer-based deep learning fat and muscle segmentation method and system, as Figure 1 shown, this method is applied to a server and includes: S101, obtaining preoperative CT image data of a target patient, gene detection information of the target patient, a preset fat and muscle segmentation model based on a deep learning network, and a training sample set.

[0016] In one implementation, the preoperative chest CT image data of the target patient is retrieved from the hospital's image database. These data cover a variety of imaging conditions, such as plain scan, enhanced scan, thin slice, thick slice, etc., to ensure a comprehensive reflection of the patient's fat and muscle conditions. For example, for a patient about to undergo abdominal surgery, the preoperative chest CT images of the patient in different scanning modes are obtained from the database, including the general morphology of the tissues clearly shown in the plain scan, the relationship between blood vessels and fat and muscle tissues more clearly presented after the enhanced scan, and the tissue details at different resolutions provided by the thin slice and thick slice scans. The format of the image data is DICOM format, which is a common standard format in the field of medical imaging and is convenient for data storage, transmission, and processing.

[0017] The blood or tissue samples of the target patient are detected using gene sequencing technology. Sample collection is the primary step in gene testing, and blood and tissue samples are important sources for obtaining gene information. The blood sample is collected by venipuncture, and 5 - 10 ml of peripheral venous blood is drawn and collected using a blood collection tube containing an anticoagulant to prevent blood clotting. Tissue samples are obtained during surgery or biopsy. For example, for abdominal surgery patients, a small amount of fat tissue or muscle tissue can be collected during surgery. The collected samples need to be processed as soon as possible to prevent nucleic acid degradation. For blood samples, components such as plasma and white blood cells are separated by methods such as centrifugation, and the DNA in them is extracted; tissue samples need to undergo a series of processes such as chopping and digestion to release the DNA in them. These extracted DNAs are the basic materials for subsequent gene sequencing.

[0018] When sequencing the genes of the target patient, first, the extracted DNA is fragmented to make it into short fragments suitable for sequencing. The length of these fragments is usually about several hundred base pairs and can be achieved by methods such as sonication or enzymatic digestion. Then, specific adapter sequences are added to both ends of the fragments. These adapters contain primer binding sites for PCR amplification and sequencing reactions. Through PCR amplification, the number of DNA fragments is enriched so that sufficient signals can be obtained in subsequent sequencing reactions. On the Illumina Hiseq platform, the sequencing reaction is based on the principle of sequencing by synthesis (SBS). During the sequencing process, four different fluorescently labeled dNTPs (deoxyribonucleotides) are added to the DNA strand being synthesized in turn according to the base complementary pairing principle. Each time a base is added, a specific color fluorescent signal is emitted, and these signals are detected by an optical system to determine the type of the added base, thereby achieving the determination of the DNA sequence. This high-throughput sequencing method can sequence a large number of DNA fragments in one experiment, greatly improving the sequencing efficiency.

[0019] The FTO gene is an important gene closely related to fat metabolism. Variations in the FTO gene are associated with an increased risk of obesity. It affects the accumulation of body fat by influencing appetite regulation, energy metabolism and other pathways. When performing genetic testing on target patients, special attention is paid to the variations at specific loci of the FTO gene. Mutations at certain loci may lead to changes in the function of the FTO protein, thereby affecting the differentiation and metabolism of adipocytes and making the body more likely to store fat. Through sequencing analysis of the FTO gene, it is possible to understand whether there are gene factors related to obesity in the patient's genetic background, which is of great significance for assessing the abnormal risk of the patient's fat content and distribution. The genes of the PPAR (peroxisome proliferator-activated receptor) family include members such as PPARα, PPARβ / δ and PPARγ, which play key roles in processes such as fat metabolism, cell differentiation and inflammatory response. PPARα is mainly expressed in tissues such as the liver and skeletal muscle, participating in the oxidative metabolism of fatty acids and regulating energy balance. PPARγ is highly expressed in adipose tissue and plays an important regulatory role in the differentiation and lipid storage of adipocytes. Detecting the expression levels and variations of PPAR family genes can provide in-depth understanding of the molecular mechanisms of the patient's fat metabolism. If certain variations occur in the PPARγ gene, it will affect its regulatory ability on adipocyte differentiation, leading to disordered fat metabolism and further affecting the patient's fat and muscle status. The MYOD1 gene is a key regulatory gene for muscle growth and development. It belongs to the myogenic regulatory factor family and can promote the differentiation of myoblasts and the formation of muscle fibers. During muscle development, the expression of the MYOD1 gene undergoes dynamic changes, regulating the expression of a series of genes related to muscle growth.

[0020] Such as Figure 2As shown, the preset fat and muscle segmentation model is constructed based on a deep learning network, using the Unet segmentation model. Its unique architecture design can effectively capture local and global features in the image, and has good adaptability for the target segmentation task of fat and muscle tissues with complex shapes and structures. The overall model adopts an Encoder-Decoder architecture. The Encoder part performs downsampling operations through multiple consecutive convolutions. Its role is to gradually extract the feature information of the image while reducing the size of the image, enabling the model to focus on the key features of the image. During this process, the number of channels of the feature map gradually increases, while the spatial resolution gradually decreases. For example, starting from the input single-channel CT image, after a series of convolution and pooling operations, the number of channels of the feature map gradually increases from the initial 1 to 64, 128, etc., while the image size is correspondingly reduced. The Decoder part then uses multiple consecutive convolutions for upsampling and restoration, gradually enlarging the feature map extracted by the Encoder to a size close to that of the original image, and at the same time fusing the features of the corresponding layer to improve the accuracy of segmentation. This method of fusing the features of the corresponding layer can make full use of the feature information at different levels, retain the details of the image, and make the segmentation result more accurate.

[0021] In the Unet model, starting from the high-level feature maps in the Encoder part, the image size is gradually restored through upsampling operations. Meanwhile, the features of the corresponding layers are concatenated and fused, and finally the segmentation result is output. The designed input of the model is a single-channel CT image because CT images can provide rich information about adipose and muscle tissues. The single-channel setting can simplify the model input and focus on the key features of CT images. The input size is (batch, 1, 512, 512), where batch refers to the amount of data required for one network iteration in network training, and in this application, it is set to 8. This means that in each training iteration, the model will simultaneously process 8 single-channel CT image data with a size of 512×512. Such a batch setting helps improve the efficiency and stability of model training. Output parameters: The output is a 4-channel (background, SMM, SAT, VAT) segmentation result, and the output size is (batch, 4, 512, 512). Each channel corresponds to the background, skeletal muscle (SMM), subcutaneous adipose tissue (SAT), and visceral adipose tissue (VAT) respectively. The numbers in the channels are the probability values belonging to that channel, and the distribution is 0~1. Finally, the result at each position is obtained by taking the maximum probability among multiple channels as the classification result at that position, forming a four-value map, thus achieving precise segmentation of different tissues. During the training process, the model continuously adjusts its own parameters to gradually reduce the value of DiceLoss to optimize the model. Network iteration optimization parameters: The cross-entropy loss is used as the network supervision method, and the network performs backpropagation learning from the errors between the prediction result and the manually annotated result through the stochastic gradient descent method, and finally converges to complete the entire learning process. In this process, the model will continuously adjust the weight parameters in the network according to the gradient information of the loss function, enabling the model to better fit the training data. Parameters such as the learning rate play a crucial role during the training process. Reasonably setting the learning rate can control the speed and effect of model convergence, and avoid problems such as overfitting or slow convergence of the model.

[0022] Patient data with complete pre - and post - operative information are selected from a large number of historical medical records accumulated in the hospital as the training sample set. These samples include patients of different ages, genders, and disease types to ensure the generalization ability of the model. For example, 1000 cases of pre - operative CT image data of patients with different diseases such as liver cancer and gastric cancer and the corresponding gene detection information are selected, including 500 male patients and 500 female patients, with the age range between 30 - 70 years old. These sample data are sorted out. The CT image data are pre - processed according to a unified standard, such as resizing the image to 512×512 pixels and normalizing the pixel values to the range of [0,1]; the gene detection information is classified and sorted out, and the mutation conditions of different gene loci are marked, etc. The sorted CT image data and gene detection information are matched one by one to form a structured training sample set for the subsequent training of the preset fat - muscle segmentation model.

[0023] S102, pre - process the pre - operative CT image data of the target patient to generate skeletal muscle image features, subcutaneous adipose tissue image features, and visceral adipose tissue image features.

[0024] In one implementation, the pre - operative CT image data of the target patient are denoised to generate a low - noise CT image. In the actual clinical environment, CT images may be interfered by various noises, such as electronic noise, quantum noise, etc. These noises will affect the image quality and the accuracy of subsequent analysis. The non - local means filtering algorithm is used to denoise the pre - operative CT image data of the target patient. This algorithm uses similar local regions in the image to estimate the true value of each pixel, thereby effectively removing noise. Taking the patient who is about to undergo abdominal tumor surgery as an example, there is obvious noise in the lung tissue and abdominal fat area of the original CT image. After non - local means filtering, the noise in the image is significantly reduced, and the originally blurred tissue boundaries become clearer. For example, the boundary between the liver edge and the surrounding adipose tissue becomes more distinct, providing a clearer basic image for subsequent processing.

[0025] Adjust the window width and window level of the low-noise CT image to generate a CT image with prominent soft tissues. According to clinical experience and the imaging characteristics of soft tissues, adjusting the window width and window level can prominently display the imaging features of specific tissues. For the low-noise CT image of this patient, set the window width to 350 - 450 HU (for example, set it to 400 HU) and the window level to 30 - 50 HU (for example, set it to 40 HU). Through such adjustments, the contrast of soft tissues in the image is enhanced, and the details of tissues such as muscles and fats become more obvious. The muscle texture of the patient's abdomen, as well as the distribution of subcutaneous fat and visceral fat, can be clearly seen. The fat layer that was not easily distinguishable under the ordinary window width and window level can clearly show its layers and scope after adjustment. Normalize the CT image with prominent soft tissues to generate a normalized CT image. To make the CT image data of different patients have a unified scale and range, facilitating model processing and comparative analysis, normalize the CT image with prominent soft tissues. Adopt the linear normalization method to map the pixel values of the image to the interval [0, 1]. After normalization, in the CT image of this patient, all pixel values are standardized to a unified range, and the differences in gray values of different tissues are more stable and comparable within this standard range. For example, regardless of the pixel value range of the original CT images of other patients, after normalization, they can be analyzed at the same scale as the images of this patient, avoiding model training biases caused by data scale differences.

[0026] Perform bone and muscle region segmentation on the normalized CT image to generate the target bone and muscle image. Use the Unet model to segment the bone and muscle region of the normalized CT image. By learning a large amount of labeled CT image data, this model can accurately identify the bone and muscle region. In the processing of this patient's image, the model successfully segmented the bones and muscles in the abdomen from other tissues, generating the target bone and muscle image. The complete contours of muscle tissues such as the rectus abdominis and external oblique abdominis in the patient's abdomen can be clearly seen, providing an accurate region for subsequent feature extraction. Extract features from the target bone and muscle image to generate bone and muscle image features. Use image feature extraction algorithms, such as the HOG (Histogram of Oriented Gradients) algorithm, to extract features from the target bone and muscle image. The HOG algorithm describes the texture and shape features of the image by calculating the histogram of gradient directions in local regions of the image. For the bone and muscle image of this patient, the HOG algorithm extracts features such as the texture direction and edge intensity of the muscles. In the rectus abdominis region, the HOG features can clearly reflect the orientation and arrangement of muscle fibers, and these features can be used to evaluate the health status and functional state of the muscles.

[0027] Perform subcutaneous adipose tissue region segmentation on the normalized CT image to generate a target subcutaneous adipose tissue image, and perform feature extraction on the target subcutaneous adipose tissue image to generate subcutaneous adipose tissue image features. The Unet model can accurately identify the range of subcutaneous adipose tissue and generate a target subcutaneous adipose tissue image. In this image, the thickness and distribution of the subcutaneous fat in the patient's abdomen can be clearly seen. Then, the gray-level co-occurrence matrix (GLCM) algorithm is used to extract features from the target subcutaneous adipose tissue image. The GLCM algorithm can calculate the spatial correlation of gray values in the image to obtain texture features. Through the GLCM algorithm, texture features of subcutaneous adipose tissue such as contrast, correlation, energy, and entropy are extracted. These features can reflect the uniformity and structural characteristics of subcutaneous adipose tissue, which is of great significance for evaluating the fat metabolism of patients.

[0028] Perform visceral adipose tissue region segmentation on the normalized CT image to generate a target visceral adipose tissue image, and perform feature extraction on the target visceral adipose tissue image to generate visceral adipose tissue image features. Taking the image data of this patient as an example, the model separates the visceral adipose tissue in the abdominal cavity from other tissues and generates a target visceral adipose tissue image. In this image, the distribution of visceral fat around organs such as the patient's liver and intestines can be clearly seen. After generating the target visceral adipose tissue image, a combination of morphological feature analysis and deep learning feature extraction is used to obtain its features. Through morphological analysis, geometric features such as area, perimeter, and circularity are calculated. For example, it is measured that the area of the visceral adipose tissue in the patient's current CT image is [X] square centimeters, the perimeter is [Y] centimeters, and the circularity value reflects the degree of regularity of its shape. These geometric features can intuitively show the scale and morphological characteristics of visceral adipose tissue, assisting in judging the degree and distribution of fat accumulation. Using a pre-trained convolutional neural network (CNN), such as ResNet, etc., to extract features from the target visceral adipose tissue image. The CNN can automatically learn complex texture and structural features in the image. The target visceral adipose tissue image is input into the pre-trained ResNet model, and the output of a specific layer is extracted as deep features. These deep features contain rich information, such as the microscopic texture of adipose tissue and its relationship with surrounding tissues.

[0029] S103. Preprocess the training sample set, adjust the window width and window level and perform normalization processing to generate target training image data.

[0030] In one implementation, the training sample set is denoised to generate low-noise CT images. During the acquisition of medical images, CT equipment is affected by various factors, resulting in noise in the CT images in the training sample set. These noises will interfere with subsequent analysis and model training. Therefore, the Gaussian filtering algorithm is used to denoise each CT image in the training sample set. By performing weighted averaging on each pixel point in the image and its neighboring pixels, the impact of noise is reduced. For the CT images of these 1000 patients, in the images of liver cancer patients, the noise will blur the boundaries between the adipose tissue and muscle tissue around the liver. After Gaussian filtering, the noise in the image is reduced, and the tissue boundaries become clearer. For example, the originally blurred boundary between the liver and the surrounding fat becomes distinct, providing a basis for subsequent image processing.

[0031] The window width and window level of the low-noise CT images are adjusted to highlight the soft tissue image features and generate normalized training image data. Due to differences in scanning equipment, scanning parameters, etc. for the CT images of different patients, the display effects of soft tissues in the images are different. To highlight the soft tissue image features and facilitate subsequent analysis of adipose and muscle tissues, according to clinical experience and the imaging characteristics of soft tissues, the window width and window level of the low-noise CT images are adjusted. The window width is set to 400 HU, and the window level is set to 40 HU. After such adjustment, in the images of these 1000 patients, the contrast of soft tissues such as muscle and fat is enhanced. The muscle texture and fat layer, which were not easily distinguishable under the ordinary window width and window level, can now be clearly displayed. For example, in the CT images of gastric cancer patients, the distribution of muscle and adipose tissues around the stomach can be observed more clearly, facilitating subsequent analysis and annotation. Then, in order to make the CT image data of different patients have a unified scale and range, facilitating model training and comparative analysis, a linear normalization method is used to process the adjusted CT images, mapping the pixel values of the images to the interval [0,1]. In this way, the CT image data of all patients are processed under the same scale, avoiding model training deviations caused by data scale differences.

[0032] Screen and label the normalized training image data, identify the skeletal muscle, subcutaneous fat tissue, and visceral fat tissue regions, and generate labeled training image data. Screen the normalized training image data to remove those images with poor quality, obvious artifacts, or incompleteness. Select 800 images with better quality from 1000 cases of data for subsequent labeling, and identify the skeletal muscle, subcutaneous fat tissue, and visceral fat tissue regions on these images. For example, during the labeling process, experts carefully outlined the contours of skeletal muscles such as the rectus abdominis and external oblique muscles in the abdomen of each image, as well as the ranges of subcutaneous fat and visceral fat, to generate labeled training image data. Perform segmentation and feature extraction on the skeletal muscle regions in the labeled training image data to generate skeletal muscle training features. From 1000 cases of labeled training image data, select the image of a liver cancer patient. Use the Unet segmentation model based on deep learning to segment the skeletal muscle region in this image. The Unet model can accurately identify the range of skeletal muscles in the patient's abdomen, such as clearly outlining the rectus abdominis and external oblique muscles, by training on a large number of image data with skeletal muscle labels. After segmentation, use the HOG algorithm to extract features from the segmented skeletal muscle images. The HOG algorithm describes the texture and shape features of an image by calculating the histogram of gradient directions in local regions of the image. On the skeletal muscle images of this patient, the HOG algorithm extracts features such as muscle texture direction and edge intensity. For example, it is found that due to liver cancer, some muscles in the patient's abdomen are atrophied, and the direction of the muscle texture has changed compared to the normal state, and the edge intensity is also different. These features are accurately captured by the HOG algorithm and form the training features of the patient's skeletal muscles.

[0033] Perform segmentation and feature extraction on the subcutaneous fat tissue region in the labeled training image data to generate subcutaneous fat tissue training features. From 1000 cases of labeled training image data, select the image of a liver cancer patient. Use other trained segmentation models to segment this training image, accurately identify the range of skeletal muscles in the patient's abdomen, such as clearly outlining the rectus abdominis and external oblique muscles. After segmentation, use the HOG algorithm to extract features from the segmented skeletal muscle images, and describe the texture and shape features of the image by calculating the histogram of gradient directions in local regions of the image. On the skeletal muscle images of this patient, the HOG algorithm extracts features such as muscle texture direction and edge intensity. For example, it is found that due to liver cancer, some muscles in the patient's abdomen are atrophied, and the direction of the muscle texture has changed compared to the normal state, and the edge intensity is also different. These features are accurately captured by the HOG algorithm and form the training features of the patient's skeletal muscles.

[0034] Similar to the processing of the skeletal muscle region, the subcutaneous adipose tissue is first accurately segmented from the labeled training image data, which helps to precisely analyze the characteristics of the subcutaneous adipose tissue. These characteristics are of great significance for understanding the patient's fat metabolism, obesity level, and the association with diseases. At the same time, they are also the key data for training the model to recognize the subcutaneous adipose tissue. Similarly, for the above-mentioned labeled training image data of liver cancer patients, a method combining threshold segmentation and morphological operations is used to segment the subcutaneous adipose tissue region. First, a suitable threshold is set according to the gray value range of the subcutaneous adipose tissue on the CT image to initially segment the subcutaneous adipose tissue. Then, through morphological operations such as dilation and erosion, some noise and small mis-segmented regions are removed to obtain a more accurate subcutaneous adipose tissue image. Then, the Gray Level Co-occurrence Matrix (GLCM) algorithm is used to extract features from the segmented subcutaneous adipose tissue image. The GLCM algorithm obtains texture features by calculating the spatial correlation of gray values in the image. From the subcutaneous adipose tissue image of this patient, the GLCM algorithm extracts features such as contrast, correlation, energy, and entropy. For example, the calculated contrast feature reflects the degree of difference between different gray regions within the subcutaneous adipose tissue. If the contrast is high, it indicates that the adipose tissue is unevenly distributed; the correlation feature reflects the similarity degree of gray values between pixels and can be used to judge whether the texture of the adipose tissue is regular; the energy feature reflects the uniformity of the gray distribution in the image, and the higher the energy value, the more regular the texture; the entropy measures the complexity of the gray distribution in the image, and a higher entropy value means that the texture of the adipose tissue is more complex.

[0035] The visceral adipose tissue region in the labeled training image data is segmented and feature extracted to generate visceral adipose tissue training features. Still taking a liver cancer patient in the selected 1000 cases of data as an example, the visceral adipose tissue region in the labeled training image data is segmented. After the segmentation is completed, a combination of morphological analysis and deep learning feature extraction is used to obtain features. Through morphological analysis, geometric features such as the area, perimeter, and circularity of the patient's visceral adipose tissue are calculated. For example, it is measured that the area of the visceral adipose tissue in the patient's current CT image is 80 square centimeters, the perimeter is 60 centimeters, and the circularity value is 0.7 (the closer the circularity is to 1, the more regular the shape). These geometric features visually show the scale and morphological characteristics of the visceral adipose tissue. The image is input into the ResNet model, and the output of a specific layer is extracted as the deep feature, which contains rich information such as the microscopic texture of the adipose tissue and the relationship with the surrounding tissues. For example, by analyzing these deep features, it is found that the boundary feature between the patient's visceral adipose tissue and the surrounding liver tissue is abnormal, suggesting the impact of the disease on the relationship between the liver and adipose tissue. These features together constitute the visceral adipose tissue training features of this patient.

[0036] Integrate the skeletal muscle training features, subcutaneous adipose tissue training features, and visceral adipose tissue training features to generate target training image data. First, arrange the feature vectors of the skeletal muscle in the front, followed by the feature vectors of the subcutaneous adipose tissue, and finally the feature vectors of the visceral adipose tissue to form a comprehensive feature vector. Perform such integration operations on each patient in 1000 cases of data. Finally, obtain the target training image data containing the comprehensive feature vectors of all patients. These data will be used as training samples and input into a preset fat and muscle segmentation model based on a deep learning network for training, enabling the model to learn the patterns and rules of various fat and muscle tissue features in different disease states. So that when processing new patient CT images subsequently, it can more accurately segment and analyze fat and muscle tissues, providing strong support for preoperative evaluation.

[0037] S104. Perform training processing on a preset fat and muscle segmentation model based on a deep learning network using the target training image data to generate a target fat and muscle segmentation model.

[0038] In one implementation, count the feature quantities of the skeletal muscle, subcutaneous adipose tissue, and visceral adipose tissue in the target training image data, and generate a sampling ratio based on the distribution balance of the above features in different tissues. After obtaining the target training image data, it is necessary to deeply analyze the respective feature quantities of the skeletal muscle, subcutaneous adipose tissue, and visceral adipose tissue. These features include various information such as texture, shape, and gray scale. Generate a sampling ratio based on the distribution balance of these features in different tissues to avoid model training biases caused by excessive or insufficient features in certain tissues and ensure that the model can learn the features of various tissues. Statistical findings show that the skeletal muscle tissue has 50 different texture features and 30 shape features; the subcutaneous adipose tissue has 40 texture features and 25 shape features; the visceral adipose tissue has 45 texture features and 28 shape features. After analysis, it is found that the texture features of the skeletal muscle tissue are relatively concentrated in the data set, while the shape features of the subcutaneous adipose tissue are more dispersed. To balance the proportions of different tissue features in subsequent training, according to the distribution of these features, the calculated sampling ratios of the skeletal muscle, subcutaneous adipose tissue, and visceral adipose tissue are 3:2:2. This ratio means that in subsequent sampling, samples will be selected from the features of different tissues according to this ratio to ensure that the model can fully learn the feature characteristics of various tissues.

[0039] Perform stratified sampling on the target training image data based on the sampling ratio, and generate a preset number of sampling feature combinations from the features of skeletal muscle, subcutaneous adipose tissue, and visceral adipose tissue. According to the sampling ratio generated previously, perform stratified sampling on the target training image data. Stratified sampling is a method of sampling according to a certain proportion from different levels or categories, ensuring that representative samples can be obtained from the features of skeletal muscle, subcutaneous adipose tissue, and visceral adipose tissue. According to the sampling ratio of 3:2:2, assuming that 100 sampling feature combinations are to be generated. Randomly select 30 groups of features from the numerous features of skeletal muscle tissue (according to the ratio 3 / (3 + 2 + 2)×100 = 30), select 20 groups from the subcutaneous adipose tissue features (2 / (3 + 2 + 2)×100 = 20), and select 20 groups from the visceral adipose tissue features. For example, in one sampling, a feature combination representing the muscle texture direction and edge sharpness is selected from the skeletal muscle tissue, a feature combination reflecting the fat distribution uniformity and texture thickness is selected from the subcutaneous adipose tissue, and a feature combination reflecting its relationship with surrounding organs and shape regularity is selected from the visceral adipose tissue. These feature combinations from different tissues are combined together to form a sampling feature combination containing information on multiple tissue features. A total of 100 such combinations are generated to prepare for subsequent analysis.

[0040] Select the features that play a key role in distinguishing skeletal muscle, subcutaneous adipose tissue, and visceral adipose tissue from all features. These key features are the core elements that can effectively reflect the differences between different tissues. Perform similarity measurement analysis on each sampling feature combination with these key features, and judge the similarity degree of each sampling feature combination with different tissues in a quantitative way. According to the analysis results, divide the data into a high-quality sample group and a low-quality sample group. The data samples in the high-quality sample group have a high similarity with the key features, can more accurately represent the characteristics of various tissues, and are of greater value for model training; the low-quality sample group is the opposite. Each sample group contains a pre-set number of data samples, and it is ensured that at least one data sample has identification information, which is used to clarify the category of the sample during the training process to help the model learn and identify different tissues. Among the 1000 cases of patient data processed previously, after in-depth analysis and screening of various features, 10 key features were determined, such as specific texture patterns (such as the fibrous texture of skeletal muscle, the uniform texture of adipose tissue), shape indices (such as the compactness of visceral adipose tissue, the distribution range index of subcutaneous adipose tissue), etc. For the 100 generated sampling feature combinations, use the cosine similarity algorithm to calculate their similarity with these 10 key features. Set the sampling feature combinations with a similarity higher than 0.7 to be divided into the high-quality sample group, and those lower than 0.5 to be divided into the low-quality sample group. For example, a sampling feature combination contains the typical fibrous texture feature and specific shape features of skeletal muscle, and its cosine similarity with the key features is 0.8, then this combination is divided into the high-quality sample group; while another sampling feature combination has less typical features and its cosine similarity with the key features is only 0.4, so it is divided into the low-quality sample group. Finally, the high-quality sample group contains 40 sampling feature combinations, and the low-quality sample group contains 60 sampling feature combinations, and at least one sample in each group has a clear identification, such as marked as "liver cancer patient - skeletal muscle", "gastric cancer patient - subcutaneous fat", etc., so that the tissue category to which the sample belongs can be clearly distinguished during model training.

[0041] Iteratively train a preset fat and muscle segmentation model based on a deep learning network using a high-quality sample group and a low-quality sample group to generate a trained model and performance metrics. During the training process, the model continuously adjusts its own parameters according to the sample data to optimize the ability to identify and segment different tissue features. In each iteration, the model attempts to reduce the difference between the prediction result and the true annotation, gradually improving the segmentation accuracy of various tissues. As the training progresses, a trained model and corresponding performance metrics are generated. These performance metrics are used to evaluate the performance of the model during training, such as accuracy, recall, Dice coefficient, etc. Through these metrics, the segmentation effect of the model on different tissues can be intuitively understood. Input the data of the divided high-quality sample group and low-quality sample group into the preset fat and muscle segmentation model based on the Unet architecture. During the training process, use the cross-entropy loss function as the optimization objective and adopt the stochastic gradient descent algorithm to adjust the weight parameters of the model. In each round of training, the model processes the sample data, calculates the loss value between the prediction result and the true annotation, and then updates the weights according to the gradient information of the loss value. After 50 rounds of training, the accuracy of the model on the training set reaches 85%, the recall is 80%, and the Dice coefficient is 0.82. These are the performance metrics of the trained model. These metrics indicate that when the model segments the fat and muscle tissues in the training data, it can accurately identify most tissue regions, and the segmentation result has a high degree of agreement with the real situation.

[0042] If the identification information in the training result is recognized by the model as the features for distinguishing skeletal muscle, subcutaneous adipose tissue, and visceral adipose tissue, and the performance metrics reach the preset threshold, then the trained model is used as the target fat and muscle segmentation model. Check whether the model in the training result can accurately identify the identification information, that is, whether the model can accurately classify the samples with identification information into the corresponding skeletal muscle, subcutaneous adipose tissue, and visceral adipose tissue categories. At the same time, compare the performance metrics of the model with the preset threshold. If the model can correctly identify the identification information and the performance metrics reach or exceed the preset threshold, it indicates that the model performs well in learning the features of different tissues and has high accuracy and reliability. This trained model can be used as the target fat and muscle segmentation model for subsequent segmentation and analysis of the fat and muscle tissues of the target patient. If the conditions are not met, the training parameters need to be adjusted or training needs to be restarted. In this training, the preset performance metric thresholds for the model are an accuracy of 80%, a recall rate of 75%, and a Dice coefficient of 0.8. After training, the recognition accuracy of the model for samples with identification information on the test set reached 83%, the recall rate was 78%, and the Dice coefficient was 0.81, all of which reached the preset threshold. At the same time, the model can accurately identify the tissue categories corresponding to different identification information. For example, for samples labeled as "liver cancer patient - skeletal muscle", the model can accurately segment them into skeletal muscle tissue, and the segmentation results are relatively consistent with the actual situation in terms of shape, position, etc. This shows that the model performs well in learning the features of different tissues and has the ability to accurately segment fat and muscle tissues. Therefore, this trained model is determined as the target fat and muscle segmentation model. In subsequent applications, this model can segment the skeletal muscle, subcutaneous adipose tissue, and visceral adipose tissue in the preoperative CT image data of the target patient, provide accurate image analysis results for doctors, and assist in preoperative evaluation. If the performance metrics of the model do not reach the preset threshold, such as an accuracy of only 75%, a recall rate of 70%, and a Dice coefficient of 0.75, the training process needs to be adjusted. Possible adjustment methods include increasing the amount of training data, reselecting key features, adjusting the hyperparameters of the model (such as the learning rate, number of iterations, etc.), or improving the architecture of the model, and then restarting the training until the model meets the preset performance requirements.

[0043] S105. Process the skeletal muscle image features, subcutaneous adipose tissue image features, and visceral adipose tissue image features based on the target fat and muscle segmentation model to generate target image segmentation information.

[0044] In one implementation, based on the target fat and muscle segmentation model, feature normalization processing is performed on the skeletal muscle image features, subcutaneous fat tissue image features, and visceral fat tissue image features to generate normalized tissue image features. For the patient undergoing abdominal surgery, the pixel value range of the skeletal muscle image features is between 0 and 255, the pixel value range of the subcutaneous fat tissue image features is between -100 and 100, and the pixel value range of the visceral fat tissue image features is between 50 and 200. Using the min-max normalization method, the pixel values of these features are all mapped to the [0, 1] interval. For the skeletal muscle image features, assuming a pixel value of 128, after normalization calculation (128 - 0) / (255 - 0) ≈ 0.5, it is transformed into the [0, 1] interval. Similar processing is also performed on the subcutaneous fat tissue and visceral fat tissue image features, so that all tissue image features are on the same scale, facilitating subsequent model processing.

[0045] Feature extraction and recognition processing are performed on the normalized tissue image features based on the target fat and muscle segmentation model to generate preliminary segmentation result features. After normalization processing, the target fat and muscle segmentation model is used to perform in-depth analysis on these image features of the unified scale. The model extracts representative features and identifies feature patterns related to skeletal muscle, subcutaneous fat tissue, and visceral fat tissue, thereby generating preliminary segmentation result features, which initially outline the possible distribution of different tissues in the image. Taking the target fat and muscle segmentation model based on the Unet architecture as an example, the convolutional layer of this model will perform convolutional operations on the normalized tissue image features to extract features such as texture, edge, and shape. When processing the patient's image, the model identifies some texture features and finds that a certain area has a specific texture pattern, similar to the texture features of the skeletal muscle tissue in the training data, so it is initially judged that this area may belong to the skeletal muscle tissue, generating corresponding preliminary segmentation result features and marking the probability information that this area may be skeletal muscle on the feature map.

[0046] Based on the target fat and muscle segmentation model, the characteristics of the preliminary segmentation results are processed for regional division to clarify the respective regions of the skeletal muscle image, subcutaneous fat tissue image, and visceral fat tissue image, generating regional division characteristics. The regions with higher probabilities in the image are merged and refined. For example, the regions with probabilities greater than 0.5 are determined as the regions of the corresponding tissues. In the image of this patient, the abdominal region is divided into the skeletal muscle region, subcutaneous fat tissue region, and visceral fat tissue region, clearly defining the boundaries of different tissues and generating regional division characteristics. These characteristics clarify the specific positions and ranges of each tissue in the image. The area calculation process is performed on the regional division characteristics to obtain the regional areas of the skeletal muscle image, subcutaneous fat tissue image, and visceral fat tissue image respectively, generating regional area characteristics. After regional division, the skeletal muscle region contains 5000 pixels. Given that the conversion relationship between the pixels of this CT image and the actual area is 1 pixel = 0.1 square centimeters. Then the regional area of the skeletal muscle image is 5000×0.1 = 500 square centimeters. Similarly, the regional area of the subcutaneous fat tissue image is calculated to be 300 square centimeters, and the regional area of the visceral fat tissue image is 200 square centimeters. These area data constitute the regional area characteristics.

[0047] The density calculation process is performed on the regional division characteristics to obtain the average densities of the skeletal muscle image, subcutaneous fat tissue image, and visceral fat tissue image respectively, generating average density characteristics. After completing the regional division, for the skeletal muscle region, assuming it contains 5000 pixels and the sum of these pixel values is 300000. Then the average density of the skeletal muscle image is 300000÷5000 = 60 (HU, Hounsfield unit, the unit of CT value). For the subcutaneous fat tissue region, if it contains 3000 pixels and the sum of the pixel values is -60000, its average density is -60000÷3000 = -20 (HU). The visceral fat tissue region contains 2000 pixels and the sum of the pixel values is 100000, and the average density is 100000÷2000 = 50 (HU). These calculated average density values constitute the average density characteristics. The average densities of different tissues vary significantly. For example, the average density of the subcutaneous fat tissue is relatively low, which conforms to the characteristics of its fat tissue, while the average densities of the skeletal muscle and visceral fat tissues are relatively high and have different value ranges. Through these average density characteristics, doctors can further understand the density conditions of different tissues in the patient's body to assist in diagnosis.

[0048] Integrate the initial segmentation result features, regional area features, and average density features to generate the target image segmentation information. Integrate the initial segmentation result features (such as the feature map marked with the possibilities of different tissues), regional area features (such as the skeletal muscle area of 500 square centimeters, subcutaneous adipose tissue area of 300 square centimeters, and visceral adipose tissue area of 200 square centimeters), and average density features (such as the average density of skeletal muscle of 60 HU, subcutaneous adipose tissue of -20 HU, and visceral adipose tissue of 50 HU). These information can be stored in a structured data form, for example, constructing a table containing fields such as tissue category, segmentation result, regional area, and average density. In this table, each row corresponds to a tissue, and the relevant information of the tissue is recorded in detail.

[0049] S106, process the target image segmentation information and the gene detection information of the target patient to generate the preoperative warning information of the target patient.

[0050] In one implementation, perform morphological analysis on the segmentation results of the skeletal muscle image, subcutaneous adipose tissue image, and visceral adipose tissue image in the target image segmentation information to generate tissue morphological feature information. For the segmentation result of the skeletal muscle image, use specialized image processing software (such as ITK-SNAP, etc.) for morphological analysis. Through the measurement tools of the software, it is found that the shape of some skeletal muscles in the patient's abdomen (such as the rectus abdominis) has changed. Normally, the rectus abdominis is in a relatively regular long strip shape, while in this patient, due to the influence of the tumor, the rectus abdominis is locally distorted and atrophied, and the shape becomes irregular. In terms of size, after software calculation, compared with healthy people of the same age and body type, the area of the rectus abdominis of the patient is significantly reduced, indicating that the muscle may have atrophied due to the disease. Looking at the edge smoothness again, the edge of normal muscle tissue is relatively smooth, while the edge of the rectus abdominis of this patient has become rough and discontinuous, which may be caused by tumor infiltration or the body's stress response. Similarly, analyzing the subcutaneous adipose tissue image, it is found that its distribution is uneven, with more fat accumulation in some areas, showing a lumpy shape, irregular shape, and different size from the normal situation, and the edge of some areas is blurred, which may be related to the patient's metabolic disorder or fat redistribution caused by the disease. For the visceral adipose tissue image, the distribution pattern around organs such as the liver and intestines has changed. The original relatively regular distribution has become disordered, the fat layer thickness in some areas has increased, there are some protrusions and depressions in the shape, and the edge is no longer smooth. These changes are all manifestations of the disease affecting fat metabolism and storage.

[0051] Statistical analysis is performed on the regional area and average density data in the target image segmentation information to generate image quantization feature information, where the image quantization feature information includes the area proportion and the density distribution difference feature. In the target image segmentation information, the regional area of the skeletal muscle image is known to be 400 square centimeters, the regional area of the subcutaneous fat tissue image is 350 square centimeters, and the regional area of the visceral fat tissue image is 250 square centimeters. By calculating the proportion of the area of each tissue in the entire abdominal area (assuming the entire abdominal area is 1000 square centimeters), the proportion of the skeletal muscle area is obtained as 40% (400÷1000×100%), the proportion of the subcutaneous fat tissue area is 35%, and the proportion of the visceral fat tissue area is 25%. Comparing with the reference data of the area proportion of abdominal fat and muscle tissues of normal people of the same age, gender, and body type, the proportion of the skeletal muscle area in normal people is usually between 45% - 55%, the proportion of the subcutaneous fat tissue area is between 20% - 30%, and the proportion of the visceral fat tissue area is between 15% - 20%. The proportion of the skeletal muscle area of this patient is lower than the normal range, while the proportions of the subcutaneous fat and visceral fat areas are relatively higher, which implies that the patient has muscle wasting and abnormal fat metabolism. From the changes in these area proportion data, it is initially judged that the body composition of the patient has become unbalanced, affecting their postoperative recovery ability and overall health status.

[0052] In terms of average density, the average density of skeletal muscle images is 50 HU (Hounsfield unit), the average density of subcutaneous adipose tissue images is -70 HU, and the average density of visceral adipose tissue images is -50 HU. Under normal circumstances, the average density of skeletal muscle is between 40 - 60 HU, the average density of subcutaneous adipose tissue is between -90 - -70 HU, and the average density of visceral adipose tissue is between -60 - -40 HU. By comparing with the average density range of normal tissues, it is found that the average density of the patient's subcutaneous adipose tissue is slightly lower than the lower limit of the normal range, indicating that there have been some changes in the composition or structure of the subcutaneous fat, such as possible changes in the lipid composition within adipocytes, or a certain degree of inflammation affecting the fat density. The average density of visceral adipose tissue is relatively higher than the normal range, which may be related to factors such as excessive fat accumulation, adipocyte hypertrophy, and local inflammatory reactions. Further analyzing the density distribution differences, the standard deviation of tissue density within different regions is calculated to evaluate the uniformity of tissue density. Suppose that in the visceral adipose tissue image of this patient, the calculated density standard deviation is relatively large, which indicates that the density distribution within the visceral adipose tissue is uneven, and there may be fat regions with different metabolic states, which is of great significance for judging the development and prognosis of the disease. Through the statistical analysis of the area and average density data of these regions, image quantification feature information including area proportion and density distribution difference characteristics is generated. If it is found that the proportion of visceral adipose area of the patient is too high and the density distribution is abnormal, it may indicate that the patient has a higher risk of cardiovascular disease, and more attention needs to be paid to cardiovascular issues during preoperative preparation and postoperative care.

[0053] In another implementation, the gene detection information of the target patient is screened for genes related to adipose muscle development and metabolism to generate key gene set information. The gene detection of this patient is completed using the Illumina Hiseq series high-throughput sequencing platform, and a large amount of gene data is detected. From these data, genes such as FTO, PPAR family (PPARα, PPARβ / δ, PPARγ), and MYOD1 are screened based on known medical research results. FTO gene mutations are associated with an increased risk of obesity and affect body fat accumulation by influencing appetite regulation and energy metabolism; PPAR family genes play a key role in fat metabolism, cell differentiation, and inflammatory responses. For example, PPARγ is crucial for the differentiation and lipid storage of adipocytes; the MYOD1 gene is a key regulatory gene for muscle growth and development. These genes constitute the key gene set information, providing core data for analyzing the genetic factors related to adipose muscle in this patient. Functional analysis is performed on the key gene set information to generate gene function impact characteristic information. For the FTO gene, it is found through analysis that a specific locus of the FTO gene in this patient has a mutation, which leads to a change in the function of the FTO protein, resulting in abnormal adipocyte differentiation and metabolism and an increased tendency of the body to store fat. In the case of the PPARγ gene, its expression level is detected to be lower than the normal range, which may affect the normal differentiation of adipocytes and the regulation of lipid storage, leading to disordered fat metabolism. The expression level of the MYOD1 gene is also abnormal, which may affect the differentiation of myoblasts and the formation of muscle fibers, resulting in poor muscle development. Combining these analyses, gene function impact characteristic information is generated, including the abnormal function of each key gene and its potential impact on the physiological processes of adipose muscle, providing a genetic basis for subsequent evaluation.

[0054] Perform correlation analysis on the tissue morphological feature information, imaging quantitative feature information, and gene function impact feature information to generate a comprehensive impact feature vector. The tissue morphological feature information shows that the abdominal muscle shape of this patient is irregular and the volume is reduced, the subcutaneous adipose tissue is unevenly distributed and thickened in some areas, and the morphological disorder of visceral adipose tissue around the liver and intestines. The imaging quantitative feature information indicates that the proportion of skeletal muscle area is lower than the normal range, the proportions of subcutaneous and visceral adipose areas are higher, and the average density of adipose tissue is also abnormal. Combining the gene function impact feature information, the mutation of the FTO gene, the abnormal expression of the PPARγ gene, and the change in the expression level of the MYOD1 gene jointly affect the normal development and metabolism of adipose muscle. For example, the mutation of the FTO gene makes the body more likely to store fat, which is related to the increase in subcutaneous and visceral adipose areas; the insufficient expression of the PPARγ gene affects lipid metabolism, further leading to abnormal lipid distribution and density changes; the abnormal expression of the MYOD1 gene is related to muscle atrophy and shape changes. Correlate and analyze this information to construct a comprehensive impact feature vector containing multiple features, such as [degree of abnormal muscle shape, proportion of muscle area, density difference of adipose tissue, impact degree of FTO gene mutation, degree of abnormal expression of PPARγ gene, degree of abnormal expression of MYOD1 gene...], and this vector comprehensively reflects the comprehensive impact of multiple factors on the adipose muscle condition of the patient.

[0055] Perform risk assessment processing on the comprehensive impact feature vector to generate risk level information related to the preoperative health status of the patient. If features such as muscle atrophy degree, abnormal distribution and density change of adipose tissue are assigned high weights in the model, and these features are relatively significant in the comprehensive impact feature vector of this patient, and at the same time, abnormal features related to the FTO gene, PPARγ gene, and MYOD1 gene also have a greater impact on the result, the model outputs a risk score after calculation. According to the preset risk level classification criteria, map the risk score to different risk levels, such as low risk, medium risk, and high risk. Assume that a risk score of 0 - 30 is low risk, 31 - 60 is medium risk, and 61 - 100 is high risk. If the risk score of this patient is 70, it is determined that the patient is at a high risk level. This means that the patient may have relatively high health risks before surgery, such as difficult postoperative recovery and an increased probability of complications. Generate preoperative warning information for the target patient based on the risk level information. For the patient determined to be at a high risk level as described above, the generated preoperative warning information may include: "Due to gene mutations (specific locus mutations of the FTO gene, abnormal expression of the PPARγ gene, change in the expression level of the MYOD1 gene) in this patient, combined with abnormal morphological and quantitative characteristics of adipose and muscle tissues (muscle atrophy, uneven adipose distribution and abnormal area ratio, adipose density change), the preoperative assessment is in a high risk state. It is recommended to further improve the examination before surgery and closely monitor the cardiovascular system function, because its disordered adipose metabolism and abnormal muscle function may increase the risk of cardiovascular diseases; in the formulation of the surgical plan, it is necessary to consider the impact of insufficient muscle strength of the patient on surgical tolerance, and reasonably adjust the anesthesia plan and surgical operation; strengthen nutritional support and rehabilitation training plans after surgery to promote muscle recovery and improve the adipose metabolism status, and reduce the probability of postoperative complications." Such preoperative warning information provides comprehensive and targeted reference for doctors, helping to improve the accuracy and effectiveness of medical decision-making.

[0056] In one implementation, as Figure 3 shown, the present application also provides a deep learning adipose and muscle segmentation device based on Slicer, including: An acquisition module 301, configured to acquire preoperative CT image data of a target patient, gene detection information of the target patient, a preset adipose and muscle segmentation model based on a deep learning network, and a training sample set; A processing module 302 is configured to preprocess the preoperative CT image data of a target patient to generate skeletal muscle image features, subcutaneous adipose tissue image features, and visceral adipose tissue image features; preprocess a training sample set, adjust the window width and window level and perform normalization processing to generate target training image data; train a preset fat and muscle segmentation model based on a deep learning network using the target training image data to generate a target fat and muscle segmentation model; process the skeletal muscle image features, subcutaneous adipose tissue image features, and visceral adipose tissue image features based on the target fat and muscle segmentation model to generate target image segmentation information; and process the target image segmentation information and the gene detection information of the target patient to generate preoperative warning information for the target patient.

[0057] The computer-readable storage medium provided in the above embodiments of the present application and the Slicer-based deep learning fat and muscle segmentation method provided in the embodiments of the present application are based on the same inventive concept and have the same beneficial effects as the methods adopted, run, or implemented by the application programs stored therein.

[0058] Each embodiment in the present application is described in a related manner. The same or similar parts among the embodiments can be referred to each other, and the differences between each embodiment and other embodiments are emphasized. In particular, for the embodiments of the Slicer-based deep learning fat and muscle segmentation method, electronic devices, electronic equipment, and readable storage media, since they are basically similar to the embodiments of the Slicer-based deep learning fat and muscle segmentation method described above, the description is relatively simple, and the relevant parts can be referred to the partial description of the embodiments of the Slicer-based deep learning fat and muscle segmentation method described above. Although the present application is disclosed as above, the present application is not limited thereto. Any person skilled in the art can make various changes and modifications without departing from the spirit and scope of the present application. Therefore, the protection scope of the present application should be defined by the scope of the claims.

Claims

1. A deep learning-based adipose and muscle segmentation method using Slicer, characterized in that, Including: Obtaining preoperative CT image data of the target patient, gene detection information of the target patient, a preset fat and muscle segmentation model based on a deep learning network, and a training sample set; Preprocessing the preoperative CT image data of the target patient to generate skeletal muscle image features, subcutaneous fat tissue image features, and visceral fat tissue image features; Preprocessing the training sample set, adjusting the window width and window level and performing normalization processing to generate target training image data; Training the preset fat and muscle segmentation model based on the deep learning network using the target training image data to generate a target fat and muscle segmentation model; Processing the skeletal muscle image features, subcutaneous fat tissue image features, and visceral fat tissue image features based on the target fat and muscle segmentation model to generate target image segmentation information, where the target image segmentation information includes the segmentation results, regional surface sum, and average density of the skeletal muscle image, subcutaneous fat tissue image, and visceral fat tissue image respectively; Processing the target image segmentation information and the gene detection information of the target patient to generate preoperative warning information for the target patient.

2. The method according to claim 1, wherein Preprocessing the preoperative CT image data of the target patient to generate skeletal muscle image features, subcutaneous fat tissue image features, and visceral fat tissue image features, including: Performing denoising processing on the preoperative CT image data of the target patient to generate a low-noise CT image; Adjusting the window width and window level of the low-noise CT image to generate a CT image with prominent soft tissues; Performing normalization processing on the CT image with prominent soft tissues to generate a normalized CT image; Performing skeletal muscle region segmentation processing on the normalized CT image to generate a target skeletal muscle image; Performing feature extraction on the target skeletal muscle image to generate skeletal muscle image features; Performing subcutaneous fat tissue region segmentation processing on the normalized CT image to generate a target subcutaneous fat tissue image; performing feature extraction on the target subcutaneous fat tissue image to generate subcutaneous fat tissue image features; Performing visceral fat tissue region segmentation processing on the normalized CT image to generate a target visceral fat tissue image; performing feature extraction on the target visceral fat tissue image to generate visceral fat tissue image features.

3. The method according to claim 1, characterized in that, Preprocessing the training sample set, adjusting the window width and window level and performing normalization processing to generate target training image data, including: Performing denoising processing on the training sample set to generate a low-noise CT image; adjusting the window width and window level of the low-noise CT image to highlight soft tissue image features and generate normalized training image data; Screening and annotating the normalized training image data to clarify the skeletal muscle, subcutaneous fat tissue, and visceral fat tissue regions and generate annotated training image data; Performing segmentation and feature extraction processing on the skeletal muscle region in the annotated training image data to generate skeletal muscle training features; Performing segmentation and feature extraction processing on the subcutaneous fat tissue region in the annotated training image data to generate subcutaneous fat tissue training features; Performing segmentation and feature extraction processing on the visceral fat tissue region in the annotated training image data to generate visceral fat tissue training features; Integrate the skeletal muscle training features, subcutaneous adipose tissue training features, and visceral adipose tissue training features to generate target training image data.

4. The method according to claim 3, wherein Perform training processing on a preset fat and muscle segmentation model based on a deep learning network using the target training image data to generate a target fat and muscle segmentation model, including: Count the feature quantities of skeletal muscle, subcutaneous adipose tissue, and visceral adipose tissue in the target training image data, and generate a sampling ratio based on the distribution balance of the above features in different tissues; Perform stratified sampling processing on the target training image data based on the sampling ratio, and generate a preset number of sampling feature combinations from the features of skeletal muscle, subcutaneous adipose tissue, and visceral adipose tissue; Select key features and perform similarity measurement analysis processing on each sampling feature combination, and divide the data into a high-quality sample group and a low-quality sample group. Among them, each sample group contains a preset number of data samples, and at least one data sample has identification information; Perform iterative training on a preset fat and muscle segmentation model based on a deep learning network using the high-quality sample group and the low-quality sample group to generate a trained model and performance indicators; If the identification information in the training result is recognized by the model as the features for distinguishing skeletal muscle, subcutaneous adipose tissue, and visceral adipose tissue, and the performance indicators reach the preset threshold, then use the trained model as the target fat and muscle segmentation model.

5. The method according to claim 1, characterized in that, Process the skeletal muscle image features, subcutaneous adipose tissue image features, and visceral adipose tissue image features based on the target fat and muscle segmentation model to generate target image segmentation information, including: Perform feature normalization processing on the skeletal muscle image features, subcutaneous adipose tissue image features, and visceral adipose tissue image features based on the target fat and muscle segmentation model to generate normalized tissue image features; Perform feature extraction and recognition processing on the normalized tissue image features based on the target fat and muscle segmentation model to generate preliminary segmentation result features; Perform region division processing on the preliminary segmentation result features based on the target fat and muscle segmentation model to clarify the respective regions of the skeletal muscle image, subcutaneous adipose tissue image, and visceral adipose tissue image, and generate region division features; Perform area calculation processing on the region division features to obtain the region areas of the skeletal muscle image, subcutaneous adipose tissue image, and visceral adipose tissue image respectively, and generate region area features; Perform density calculation processing on the region division features to obtain the average densities of the skeletal muscle image, subcutaneous adipose tissue image, and visceral adipose tissue image respectively, and generate average density features; Integrate the preliminary segmentation result features, region area features, and average density features to generate target image segmentation information.

6. The method according to claim 1, wherein Process the target image segmentation information and the gene detection information of the target patient to generate preoperative warning information for the target patient, including Perform morphological analysis processing on the respective segmentation results of the skeletal muscle image, subcutaneous adipose tissue image, and visceral adipose tissue image in the target image segmentation information to generate tissue morphological feature information. Among them, the tissue morphological feature information includes tissue shape, size, and edge smoothness features; Statistical analysis is performed on the regional area and average density data in the target image segmentation information to generate image quantization feature information, where the image quantization feature information includes area ratio and density distribution difference features.

7. The method according to claim 6, wherein The target image segmentation information and the gene detection information of the target patient are processed to generate preoperative warning information for the target patient, further including: Screening and processing the gene detection information of the target patient for genes related to fat and muscle development and metabolism to generate key gene set information; Performing functional analysis on the key gene set information to generate gene function influence feature information; Performing correlation analysis on the tissue morphological feature information, image quantization feature information, and gene function influence feature information to generate a comprehensive influence feature vector; Performing risk assessment on the comprehensive influence feature vector to generate risk level information related to the preoperative health status of the patient; generating preoperative warning information for the target patient based on the risk level information.

8. A deep learning fat and muscle segmentation device based on Slicer, characterized in that, The device includes: An acquisition module, configured to acquire preoperative CT image data of the target patient, gene detection information of the target patient, a preset fat and muscle segmentation model based on a deep learning network, and a training sample set; A processing module, configured to preprocess the preoperative CT image data of the target patient to generate skeletal muscle image features, subcutaneous adipose tissue image features, and visceral adipose tissue image features; preprocess the training sample set, adjust the window width and window level and perform normalization processing to generate target training image data; train the preset fat and muscle segmentation model based on the deep learning network using the target training image data to generate a target fat and muscle segmentation model; process the skeletal muscle image features, subcutaneous adipose tissue image features, and visceral adipose tissue image features based on the target fat and muscle segmentation model to generate target image segmentation information; process the target image segmentation information and the gene detection information of the target patient to generate preoperative warning information for the target patient.

9. An electronic device, characterized in that, Including: A first processor; And a memory, configured to store executable instructions of the first processor; wherein, the first processor is configured to execute the method for deep learning fat and muscle segmentation based on Slicer according to any one of claims 1 to 7 by executing the executable instructions.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a second processor, it implements the method for deep learning fat and muscle segmentation based on Slicer according to any one of claims 1 to 7.

Images