Deep learning-based multi-organ segmentation system and colon cancer detection system

Abstract

The application discloses a kind of multi-organ segmentation systems and colon cancer detection systems based on deep learning, multi-organ segmentation system includes: image processing module, for converting CT image into NIFTI three-dimensional image format, carries out data standardization processing;Again divide several sub three-dimensional images;Multi-organ detection segmentation module is used to the spleen, right kidney, left kidney, liver, aorta, inferior vena cava region in CT image is segmented.The colon cancer detection system includes: image processing module;Detection segmentation module is used to the colon cancer region in PET-CT image is segmented;Quantitative analysis module is used to the standard uptake value of colon cancer region is quantitatively analyzed.The application realizes the segmentation of each region in CT image;While the colon cancer lesion in PET-CT image is segmented, the standard uptake value of lesion region is quantitatively analyzed;Give consideration to the simplicity of operation, the accuracy of segmentation result and quantitative analysis function, realize the accurate segmentation and automatic analysis function of colon cancer region.

Description

Technical Field

[0001] This invention belongs to the field of digital organ segmentation and colorectal cancer detection and analysis, and particularly relates to a deep learning-based multi-organ segmentation system and a colorectal cancer detection system. Background Technology

[0002] Medical imaging plays a crucial role in the diagnosis, treatment, and monitoring of various diseases. Three-dimensional medical imaging technologies such as positron emission tomography (PET-CT) and computed tomography (CT) are widely used due to their ability to provide detailed images of internal structures. The goal of medical image segmentation is to identify and quantify specific regions in images of biological tissues or organs that play a critical role in disease diagnosis, radiotherapy, and preoperative planning. Accurate segmentation of organs in CT images allows physicians to obtain detailed information about organ morphology, structure, and function, thus aiding in early disease diagnosis and quantitative assessment. Before surgery, organ segmentation provides the anatomical information needed for surgical navigation and planning, helping physicians assess surgical feasibility, determine surgical pathways, and predict surgical outcomes. Colorectal cancer is a common malignant tumor, and PET-CT plays a vital role in its diagnosis and treatment. Segmentation of PET-CT images of colorectal cancer allows for accurate separation of the cancer from surrounding tissue and background. This helps determine the tumor's boundaries and morphological characteristics, providing a basis for further analysis and diagnosis. Furthermore, PET-CT combines the metabolic information from PET with the anatomical information from CT, providing a more comprehensive detection of colorectal cancer. PET scans can reveal the metabolic activity of tumors, while CT scans can provide information on the morphology and location of tumors. Combining the two can enhance the detection and localization capabilities for colorectal cancer.

[0003] Due to the large volume and complex structure of medical image data, manual medical image segmentation is a time-consuming and arduous task for doctors, and their subjective judgment may lead to a certain degree of error and inconsistency. Traditional image segmentation methods, such as Bayesian classifiers, graph cuts, and watershed transforms, typically rely on predefined rules and assumptions about the data. However, these methods are limited in their flexibility and generalization ability in complex medical image tasks. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies by providing a deep learning-based multi-organ segmentation system and a colorectal cancer detection system and method.

[0005] The objective of this invention is achieved through the following technical solution: a multi-organ segmentation system based on deep learning, comprising:

[0006] The image processing module is used to convert CT images into NIFTI 3D image format, perform data standardization processing, and then divide them into several sub-3D images by sequential block cutting.

[0007] The multi-organ detection and segmentation module is used to input sub-3D images into a deep learning model to segment the spleen, right kidney, left kidney, liver, aorta, and inferior vena cava regions in CT images;

[0008] The deep learning model consists of an encoder and a decoder, with a total of six layers. Both the encoder and decoder are composed of pure convolutional modules and parallel hybrid modules. The first two layers of the encoder use pure convolutional modules and downsampling operations to gradually reduce the spatial size. The decoder uses pure convolutional modules to restore the original image size through upsampling. The parallel hybrid modules in the last four layers of the deep learning model are composed of 3D Shunted Transformer and CNN. The encoder and decoder employ cross-layer connections to improve the accuracy of segmentation.

[0009] Furthermore, the decoder also introduces deep supervision to enhance the directness and transparency of the hidden layer learning process.

[0010] Furthermore, the data standardization process specifically involves applying Z-Score standardization to the 3D CT images. First, the 0.5% and 99.5% quantile pixel values ​​of the 3D CT images are statistically analyzed, and values ​​outside these ranges are removed to eliminate the influence of outliers. Then, Z-Score standardization is applied, calculated using the following formula:

[0011]

[0012] Where x represents the pixel value of a certain pixel in the image, x ′ This represents the pixel value after normalization, in μ. * δ represents the mean of pixel values ​​between the 0.5th and 99.5th percentiles. * This represents the standard deviation of pixel values ​​between the 0.5th percentile and the 99.5th percentile.

[0013] Furthermore, the step of inputting the sub-3D images into the deep learning model to segment the spleen, right kidney, left kidney, liver, aorta, and inferior vena cava regions in the CT image specifically involves: inputting the sub-3D images into the deep learning model to obtain the segmentation results on the sub-3D images, i.e., obtaining the sub-predicted 3D images; and finally, stitching together the segmentation results of several sub-predicted 3D images to obtain the segmentation results of the spleen, right kidney, left kidney, liver, aorta, and inferior vena cava regions in the entire 3D image.

[0014] A deep learning-based colorectal cancer detection system includes:

[0015] The image processing module is used to register PET-CT images of two-dimensional DICOM sequences, aligning the PET images with the CT images as the standard; then, the registered PET-CT images are converted into NIFTI three-dimensional image format and subjected to data standardization; finally, several sub-three-dimensional images are divided into several sub-three-dimensional images by sequentially slicing the data-standardized three-dimensional images.

[0016] The detection and segmentation module is used to input the sub-3D image into the deep learning model to segment the colon cancer region in the PET-CT image;

[0017] The quantitative analysis module is used to quantitatively analyze the standard uptake values ​​of colorectal cancer regions to reflect the metabolic level and activity of the lesions;

[0018] The deep learning model consists of six layers, comprising an encoder and a decoder. Both the encoder and decoder are composed of pure convolutional modules and parallel hybrid modules. Pure convolutional blocks and downsampling operations are introduced in the first two layers of the encoder to gradually reduce the spatial size. The decoder uses pure convolutional modules to recover the original image size through upsampling. The parallel hybrid modules in the last four layers of the deep learning model consist of modified 3D versions of the Shunted Transformer and CNN. Cross-layer connections are used between the encoder and decoder to improve the accuracy of segmentation.

[0019] Furthermore, the decoder also introduces deep supervision to enhance the directness and transparency of the hidden layer learning process.

[0020] Furthermore, the data standardization process specifically involves standardizing the PET images using the Z-Score method, with the calculation formula as follows:

[0021]

[0022] Where x represents the pixel value of a certain pixel in the image, x ′ δ represents the pixel value after normalization, μ represents the mean of the pixel values ​​in the 3D PET image, and δ represents the standard deviation of the pixel values ​​in the 3D PET image.

[0023] Furthermore, the step of inputting the sub-3D image into the deep learning model to segment the colon cancer region in the image specifically involves: inputting the sub-3D image into the deep learning model to obtain the segmentation result on the sub-3D image, i.e., obtaining the sub-predicted 3D image; and stitching together several sub-predicted 3D images to obtain the segmentation result of the colon cancer region in the entire 3D image.

[0024] Furthermore, the PET-CT image registration specifically involves: firstly, selecting the Mattes mutual information metric to evaluate the similarity between the PET image and the CT image; then, employing a gradient descent optimization strategy to find the optimal registration parameters; using B-spline interpolation to obtain pixel values ​​at non-integer locations; subsequently, performing image registration; and finally, after registration is complete, applying the transformation to the PET image to obtain the registered PET image, thereby obtaining the registered PET-CT image.

[0025] Further, the quantitative analysis of the standard uptake value for the colorectal cancer region specifically involves: extracting patient information obtained from the PET scan from the DICOM header file of the PET scanner, including the injection dose, patient weight, half-life of the radiopharmaceutical, radiopharmaceutical injection start time, and scan start time; then calculating the scan time from radiopharmaceutical injection to the PET scan; then calculating the radiopharmaceutical decay correction factor using the half-life and scan time; then calculating the standard uptake factor using the patient weight, decay correction factor, and injection dose; finally, multiplying the pixel values ​​of the PET image of the colorectal cancer region by the standard uptake factor to obtain the standard uptake value for that region.

[0026] The formula for calculating the decay correction factor of a radiopharmaceutical is as follows:

[0027]

[0028] Where scan_time_seconds is the time from the injection of the radiopharmaceutical to the PET scan; half_life is the half-life of the injected radiopharmaceutical.

[0029] The formula for calculating the standard intake factor is:

[0030]

[0031] Here, patient_weight represents the patient's weight; decay_factor represents the decay correction factor of the radiopharmaceutical; and injected_dose represents the total dose of radiopharmaceutical injected into the patient.

[0032] The beneficial effects of this invention are:

[0033] (1) It achieves accurate segmentation of the spleen, right kidney, left kidney, liver, aorta and inferior vena cava regions in CT images; while accurately segmenting colon cancer lesions in PET-CT images, it quantitatively analyzes the standard uptake value (SUV) of the segmented lesion regions.

[0034] (2) It takes into account the simplicity of the operation process, the accuracy of the segmentation results and the quantitative analysis function, and realizes the precise segmentation of the colon cancer region and the automatic analysis function of colon cancer detection. Attached Figure Description

[0035] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0036] Figure 1 This is an overall flowchart of the multi-organ segmentation system of the present invention;

[0037] Figure 2 This is a segmentation diagram of the spleen using the multi-organ segmentation system of the present invention;

[0038] Figure 3 This is a segmentation diagram of the right kidney using the multi-organ segmentation system of the present invention;

[0039] Figure 4 This is a segmentation diagram of the left kidney using the multi-organ segmentation system of the present invention;

[0040] Figure 5 This is a segmentation diagram of the liver using the multi-organ segmentation system of the present invention;

[0041] Figure 6 This is a segmentation diagram of the aorta using the multi-organ segmentation system of the present invention;

[0042] Figure 7 This is a segmentation diagram of the inferior vena cava using the multi-organ segmentation system of the present invention;

[0043] Figure 8 This is the result of quantitative analysis of the segmentation map and standard uptake values ​​of the segmented regions of colon cancer by the colon cancer detection system of the present invention;

[0044] Figure 9 This is a structural diagram of the deep learning model of the present invention;

[0045] Figure 10 This is an overall flowchart of the colon cancer detection system of the present invention. Detailed Implementation

[0046] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Rather, they are merely examples of apparatuses and methods consistent with some aspects of the invention as detailed in the appended claims.

[0047] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The singular forms “a,” “the,” and “the” used in this invention and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any or all possible combinations of one or more of the associated listed items.

[0048] It should be understood that although the terms first, second, third, etc., may be used in this invention to describe various information, this information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, first information may also be referred to as second information without departing from the scope of this invention, and similarly, second information may also be referred to as first information. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to a determination."

[0049] The present invention will now be described in detail with reference to the accompanying drawings. Unless otherwise specified, the features of the following embodiments and implementations can be combined with each other.

[0050] This invention provides a deep learning-based multi-organ segmentation system, such as... Figure 1 As shown, it includes an image processing module and a multi-organ detection and segmentation module.

[0051] The image processing module converts CT images into NIFTI 3D image format and performs data standardization. It then divides the images into several sub-3D images using a sequential block-slicing method. Specifically, the data standardization process involves applying Z-Score standardization to the 3D CT images. First, the 0.5% and 99.5% quantile pixel values ​​of the 3D CT images are statistically analyzed, and values ​​outside these ranges are removed to eliminate the influence of outliers. Then, Z-Score standardization is applied, calculated using the following formula:

[0052]

[0053] Where x represents the pixel value of a certain pixel in the image, x′ represents the pixel value of that pixel after normalization, and μ *δ represents the mean of pixel values ​​between the 0.5th and 99.5th percentiles. * This represents the standard deviation of pixel values ​​between the 0.5th percentile and the 99.5th percentile.

[0054] The multi-organ detection and segmentation module is used to input sub-3D images into a deep learning model to segment the spleen, right kidney, left kidney, liver, aorta, and inferior vena cava regions in CT images;

[0055] The specific implementation process is as follows:

[0056] (1) Convert DICOM format 2D CT image sequences to NIFTI format 3D image format;

[0057] The format conversion can be achieved by reading the DICOM image sequence in the folder using the SimpleITK.ImageSeriesReader() function in SimpleITK, and then saving the read DICOM sequence as an NIFTI image format using the SimpleITK.WriteImage() function.

[0058] (2) Standardize the NIFTI format CT image data;

[0059] Standardizing image data ensures rapid convergence of deep learning models during training, improving computational efficiency. Z-Score standardization is employed by first statistically analyzing the 0.5% and 99.5% quantile pixel values ​​of the 3D CT image and discarding values ​​outside these ranges to eliminate outliers. Then, Z-Score standardization is applied using the following formula:

[0060]

[0061] Where x represents the pixel value of a certain pixel in the image, x^' represents the pixel value of that pixel after normalization, μ^* represents the mean of pixel values ​​between the 0.5% quantile and the 99.5% quantile, and δ^* represents the standard deviation of pixel values ​​between the 0.5% quantile and the 99.5% quantile.

[0062] (3) Perform sequential block cutting on the standardized three-dimensional image;

[0063] The size of the sub-3D image after slicing is (48, 192, 192) in this embodiment.

[0064] (4) Input the segmented sub-3D images into the deep learning model to obtain the segmentation results of the specified organ region in the sub-3D images, that is, to obtain the sub-predicted 3D images; stitch together the segmentation results of several sub-predicted 3D images to obtain the segmentation results of the colon cancer region in the entire 3D image.

[0065] like Figures 2-7 As shown, multiple sub-predicted 3D images are sequentially stitched together to obtain the spleen in the entire 3D image. Figure 2 (Middle black diagonal area), right kidney () Figure 3 (Middle black diagonal area), left kidney ( Figure 4 (Black diagonal area), liver () Figure 5 (Black diagonal area), aorta ( Figure 6 (Black diagonal line area), inferior vena cava ( Figure 7 The segmentation result of the region (the area marked with black diagonal lines).

[0066] like Figure 9 As shown, the deep learning model is named MPSTrans and consists of six layers. In the construction of MPSTrans, the encoder and decoder adopt a U-shaped design, composed of pure convolutional modules and parallel hybrid modules. Considering the computational complexity of the self-attention mechanism in Transformers, directly processing pixel-labeled input is not feasible. Therefore, pure convolutional blocks and downsampling operations are introduced in the first two layers of the encoder to gradually reduce the spatial size. This approach helps extract low-level features at high resolution while capturing fine-grained spatial information. Similarly, the decoder uses pure convolutional modules at the same stage to recover the original image size through upsampling. Furthermore, the parallel hybrid modules in the last four layers of the model mainly consist of modified 3D versions of the Shunted Transformer and CNN, which can simultaneously aggregate local and global semantic information. Cross-layer connections are used between the encoder and decoder (the feature information extracted by the encoder layer is directly input into the decoder layer) to improve segmentation accuracy. In addition, deep supervision is introduced in the decoder to enhance the directness and transparency of the hidden layer learning process.

[0067] Before using the deep learning model, it is trained and tested. Specifically, the following steps are taken: training and testing sets are divided from CT images; pixel-level annotations are performed on the training and testing sets in the CT images, so that the background, spleen, right kidney, left kidney, liver, aorta, and inferior vena cava regions have different label values; pixel-level annotations are also performed on the training and testing sets in the PET-CT images, so that the background and colon cancer regions have different label values; the annotated training set is used to train the deep learning model, and the annotated testing set is used to test the deep learning model, thus obtaining the deep learning model.

[0068] This invention provides a deep learning-based colorectal cancer detection system, such as... Figure 10 As shown, it includes an image processing module, a detection and segmentation module, and a quantitative analysis module.

[0069] The image processing module is used to register PET-CT images of two-dimensional DICOM sequences, aligning the PET images with the CT images as the standard; then, the registered PET-CT images are converted into NIFTI three-dimensional image format and subjected to data standardization; finally, several sub-three-dimensional images are divided into several sub-three-dimensional images by sequentially slicing the data-standardized three-dimensional images.

[0070] The PET-CT image registration process is as follows: First, the Mattes mutual information metric is selected to evaluate the similarity between the PET image and the CT image; then, a gradient descent optimization strategy is used to find the optimal registration parameters; B-spline interpolation is used to obtain pixel values ​​at non-integer locations; then, image registration is performed; finally, after registration is completed, the transformation is applied to the PET image to obtain the registered PET image, thus obtaining the registered PET-CT image.

[0071] The PET images were standardized using the Z-Score method, and the calculation formula is as follows:

[0072]

[0073] Where x represents the pixel value of a certain pixel in the image, x ′ δ represents the pixel value after normalization, μ represents the mean of the pixel values ​​in the 3D PET image, and δ represents the standard deviation of the pixel values ​​in the 3D PET image.

[0074] The detection and segmentation module is used to input the sub-3D image into the deep learning model to segment the colon cancer region in the PET-CT image;

[0075] Specifically, the process involves inputting the sub-3D image into a deep learning model to obtain the segmentation result on the sub-3D image, which is the sub-predicted 3D image; and then stitching together multiple sub-predicted 3D images to obtain the segmentation result of the colon cancer region in the entire 3D image.

[0076] The quantitative analysis module is used to quantitatively analyze the standard uptake values ​​of colorectal cancer regions to reflect the metabolic level and activity of the lesions;

[0077] Specifically, the process involves: extracting patient information obtained from a PET scan from the DICOM header file of the PET scanner, including the injection dose, patient weight, half-life of the radiopharmaceutical, radiopharmaceutical injection start time, and scan start time; calculating the scan time from radiopharmaceutical injection to the PET scan; calculating the radiopharmaceutical decay correction factor using the half-life and scan time; calculating the standard uptake factor using the patient weight, decay correction factor, and injection dose; and finally, multiplying the pixel values ​​of the PET image of the colon cancer region by the standard uptake factor to obtain the standard uptake value for that region.

[0078] The formula for calculating the decay correction factor of a radiopharmaceutical is as follows:

[0079]

[0080] Where scan_time_seconds is the time from the injection of the radiopharmaceutical to the PET scan; half_life is the half-life of the injected radiopharmaceutical.

[0081] The formula for calculating the standard intake factor is:

[0082]

[0083] Here, patient_weight represents the patient's weight; decay_factor represents the decay correction factor of the radiopharmaceutical; and injected_dose represents the total dose of radiopharmaceutical injected into the patient.

[0084] Before using the deep learning model, the deep learning model is trained and tested. Specifically, the training set and the test set are divided from the PET-CT images; the training set and the test set in the PET-CT images are labeled at the pixel level so that the background and the colorectal cancer area have different label values; the labeled training set is used to train the deep learning model, and the labeled test set is used to test the deep learning model, thus obtaining the deep learning model.

[0085] like Figure 9As shown, the deep learning model is named MPSTrans and consists of six layers. In the construction of MPSTrans, the encoder and decoder adopt a U-shaped design, composed of pure convolutional modules and parallel hybrid modules. Considering the computational complexity of the self-attention mechanism in Transformers, directly processing pixel-labeled input is not feasible. Therefore, pure convolutional blocks and downsampling operations are introduced in the first two layers of the encoder to gradually reduce the spatial size. This approach helps extract low-level features at high resolution while capturing fine-grained spatial information. Similarly, the decoder uses pure convolutional modules at the same stage to recover the original image size through upsampling. Furthermore, the parallel hybrid modules in the last four layers of the model mainly consist of modified 3D versions of the Shunted Transformer and CNN, which can simultaneously aggregate local and global semantic information. Cross-layer connections are used between the encoder and decoder (the feature information extracted by the encoder layer is directly input into the decoder layer) to improve segmentation accuracy. In addition, deep supervision is introduced in the decoder to enhance the directness and transparency of the hidden layer learning process.

[0086] The process of implementing a colon cancer detection system is as follows:

[0087] (1) Register PET images in DICOM sequence format with CT images;

[0088] The PET-CT image registration process is as follows: First, the Mattes mutual information metric is selected to evaluate the similarity between the PET image and the CT image; then, a gradient descent optimization strategy is used to find the optimal registration parameters; B-spline interpolation is used to obtain pixel values ​​at non-integer positions; after setting the above methods, image registration is performed; finally, after registration is completed, the transformation is applied to the PET image to obtain the registered PET image.

[0089] (2) Convert the registered DICOM sequence format PET images and CT images to NIFTI image format;

[0090] (3) Standardize the data of PET-CT images in NIFTI format;

[0091] The data standardization process for CT images first involves statistically analyzing the 0.5% and 99.5% quantile pixel values ​​of the 3D CT image and removing values ​​outside these ranges to eliminate the influence of outliers; then, Z-score standardization is applied, calculated using the following formula:

[0092]

[0093] Where x represents the pixel value of a certain pixel in the image, x^' represents the pixel value of that pixel after normalization, μ^* represents the mean of pixel values ​​between the 0.5% quantile and the 99.5% quantile, and δ^* represents the standard deviation of pixel values ​​between the 0.5% quantile and the 99.5% quantile.

[0094] The data standardization process involves directly applying the Z-Score method to PET images, with the calculation formula as follows:

[0095]

[0096] Where x represents the pixel value of a certain pixel in the image, x^' represents the pixel value of that pixel after normalization, μ represents the mean of the pixel values ​​in the 3D PET image, and δ represents the standard deviation of the pixel values ​​in the 3D PET image.

[0097] (4) Perform sequential block cutting on the standardized three-dimensional image;

[0098] In this embodiment, the size of the sub-3D image after slicing is (64, 160, 160).

[0099] (5) Input the segmented sub-3D images into the deep learning model to obtain the segmentation results of the colon cancer region in the sub-3D images, that is, to obtain the sub-predicted 3D images; stitch together several sub-predicted 3D images to obtain the segmentation results of the colon cancer region in the entire 3D image;

[0100] To ensure the accuracy of cell segmentation results, the deep learning model also needs to be trained and tested. Specifically, the training set and test set are divided from the PET-CT images in an 8:2 ratio; the training set and test set in the PET-CT images are labeled at the pixel level so that the background and the colon cancer area have different label values; the labeled training set is used to train the deep learning model, and the labeled test set is used to test the deep learning model, thus obtaining the deep learning model.

[0101] The deep learning model has the same structure as the model used in the multi-organ segmentation module.

[0102] (6) Quantitative analysis was performed on the colon cancer region to obtain the mean quasi-uptake value (avg_suv), maximum quasi-uptake value (max_suv), and minimum standard uptake value (min_suv) for the colon cancer region, such as... Figure 8 As shown.

[0103] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

[0104] The above embodiments are only used to illustrate the design concept and features of the present invention, and their purpose is to enable those skilled in the art to understand the content of the present invention and implement it accordingly. The protection scope of the present invention is not limited to the above embodiments. Therefore, all equivalent changes or modifications made based on the principles and design ideas disclosed in the present invention are within the protection scope of the present invention.

[0105] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the disclosure herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein. The specification and embodiments are to be considered exemplary only.

[0106] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope.

Claims

1. A multi-organ segmentation system based on deep learning, characterized in that, include: The image processing module is used to convert CT images into NIFTI 3D image format and perform data standardization processing; Then, several sub-3D images are divided into segments using a sequential slicing method; The multi-organ detection and segmentation module is used to input sub-3D images into a deep learning model to segment the spleen, right kidney, left kidney, liver, aorta, and inferior vena cava regions in CT images; The deep learning model consists of an encoder and a decoder, with a total of six layers. Both the encoder and decoder are composed of pure convolutional modules and parallel hybrid modules. The first two layers of the encoder use pure convolutional modules and downsampling operations to gradually reduce the spatial size. The decoder uses pure convolutional modules to restore the original image size through upsampling. The parallel hybrid modules in the last four layers of the deep learning model are composed of 3D Shunted Transformer and CNN. The encoder and decoder employ cross-layer connections to improve the accuracy of segmentation.

2. The deep learning-based multi-organ segmentation system according to claim 1, characterized in that, The decoder also introduces deep supervision to enhance the directness and transparency of the hidden layer learning process.

3. The deep learning-based multi-organ segmentation system according to claim 1, characterized in that, The data standardization process specifically involves applying Z-Score standardization to the 3D CT images. First, the 0.5% and 99.5% quantile pixel values ​​of the 3D CT images are statistically analyzed, and values ​​outside these ranges are removed to eliminate the influence of outliers. Then, Z-Score standardization is applied, calculated using the following formula: Where x represents the pixel value of a certain pixel in the image, x′ represents the pixel value of that pixel after normalization, and μ * δ represents the mean of pixel values ​​between the 0.5th and 99.5th percentiles. * This represents the standard deviation of pixel values ​​between the 0.5th percentile and the 99.5th percentile.

4. The deep learning-based multi-organ segmentation system according to claim 1, characterized in that, The specific steps for inputting the sub-3D images into the deep learning model to segment the spleen, right kidney, left kidney, liver, aorta, and inferior vena cava regions in the CT image are as follows: inputting the sub-3D images into the deep learning model to obtain the segmentation results on the sub-3D images, i.e., obtaining the sub-predicted 3D images; and finally, stitching together the segmentation results of several sub-predicted 3D images to obtain the segmentation results of the spleen, right kidney, left kidney, liver, aorta, and inferior vena cava regions in the entire 3D image.

5. A deep learning-based colorectal cancer detection system, characterized in that, include: The image processing module is used to register PET-CT images of two-dimensional DICOM sequences, aligning the PET images with the CT images as the standard. The registered PET-CT images are then converted to NIFTI 3D image format and then subjected to data standardization. Then, the three-dimensional image, after data standardization, is divided into several sub-three-dimensional images by sequential slicing. The detection and segmentation module is used to input sub-3D images into a deep learning model to segment the colon cancer region in PET-CT images; The quantitative analysis module is used to quantitatively analyze the standard uptake values ​​of colorectal cancer regions to reflect the metabolic level and activity of the lesions; The deep learning model consists of six layers, comprising an encoder and a decoder. Both the encoder and decoder are composed of pure convolutional modules and parallel hybrid modules. Pure convolutional blocks and downsampling operations are introduced in the first two layers of the encoder to gradually reduce the spatial size. The decoder uses pure convolutional modules to recover the original image size through upsampling. The parallel hybrid modules in the last four layers of the deep learning model consist of modified 3D versions of the Shunted Transformer and CNN. Cross-layer connections are used between the encoder and decoder to improve the accuracy of segmentation.

6. A deep learning-based colon cancer detection system according to claim 5, characterized in that, The decoder also introduces deep supervision to enhance the directness and transparency of the hidden layer learning process.

7. A deep learning-based colon cancer detection system according to claim 5, characterized in that, The data standardization process specifically involves applying the Z-Score method to the PET images for standardization, with the following calculation formula: Where x represents the pixel value of a certain pixel in the image, x ′ δ represents the pixel value after normalization, μ represents the mean of the pixel values ​​in the 3D PET image, and δ represents the standard deviation of the pixel values ​​in the 3D PET image.

8. A deep learning-based colon cancer detection system according to claim 5, characterized in that, The step of inputting the sub-3D image into the deep learning model to segment the colon cancer region in the image specifically involves: inputting the sub-3D image into the deep learning model to obtain the segmentation result on the sub-3D image, i.e., obtaining the sub-predicted 3D image; and stitching together several sub-predicted 3D images to obtain the segmentation result of the colon cancer region in the entire 3D image.

9. A deep learning-based colon cancer detection system according to claim 5, characterized in that, The PET-CT image registration process is as follows: First, the Mattes mutual information metric is selected to evaluate the similarity between the PET image and the CT image; then, a gradient descent optimization strategy is used to find the optimal registration parameters; B-spline interpolation is used to obtain pixel values ​​at non-integer locations; then, image registration is performed; finally, after registration is completed, the transformation is applied to the PET image to obtain the registered PET image, thus obtaining the registered PET-CT image.

10. A deep learning-based colorectal cancer detection system according to claim 5, characterized in that, The quantitative analysis of the standard uptake value for the colorectal cancer region specifically involves: extracting patient information obtained from the PET scan from the DICOM header file of the PET scanner, including the injection dose, patient weight, half-life of the radiopharmaceutical, radiopharmaceutical injection start time, and scan start time; then calculating the scan time from radiopharmaceutical injection to the PET scan; next, calculating the radiopharmaceutical decay correction factor using the half-life and scan time; then calculating the standard uptake factor using the patient weight, decay correction factor, and injection dose; finally, multiplying the pixel values ​​of the PET image of the colorectal cancer region by the standard uptake factor to obtain the standard uptake value for that region. The formula for calculating the decay correction factor of a radiopharmaceutical is as follows: Where scan_time_seconds is the time from the injection of the radiopharmaceutical to the PET scan; half_life is the half-life of the injected radiopharmaceutical. The formula for calculating the standard intake factor is: Here, patient_weight represents the patient's weight; decay_factor represents the decay correction factor of the radiopharmaceutical; and injected_dose represents the total dose of radiopharmaceutical injected into the patient.

Images