Colon cancer CT image segmentation method and system fusing individualized features

Abstract

The invention belongs to the field of medical image processing and artificial intelligence, and provides a colon cancer CT image segmentation method and system fusing individualized features, and the method comprises the steps: obtaining the abdomen CT images of a historical patient and a target patient; obtaining structured individual feature information related to the patient, wherein the individual feature information comprises age, gender, colon cancer family history and intestinal medical history information; preprocessing the abdomen CT image; and training an image segmentation model by using the preprocessed abdominal CT image of the historical patient, segmenting the preprocessed abdominal CT image of the target patient by using the trained image segmentation model meeting the segmentation precision requirement, and generating a colon cancer focus segmentation image conforming to the individual feature difference of the target patient. According to the method, the structured patient information is utilized to guide the model network to adjust the feature response, so that the segmentation precision and individual adaptability of the model are improved, and the segmentation accuracy is further improved.

Description

Technical Field

[0001] This invention belongs to the field of medical image processing and artificial intelligence, and specifically relates to a method and system for segmenting colorectal cancer CT images that integrates individualized features. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] Early detection and accurate diagnosis are crucial for improving the prognosis of colorectal cancer patients. Currently, clinical practice mainly relies on methods such as abdominal computed tomography (CT) and colonoscopy for colorectal cancer screening and diagnosis.

[0004] In CT image analysis, lesion segmentation is a crucial foundation for assisting physicians in tumor localization, size assessment, and surgical planning. Traditional segmentation methods often rely on manual delineation or image processing algorithms based on rules such as thresholding, region growing, and edge detection. These methods have the following drawbacks: they depend on manual operation, resulting in low efficiency and high subjectivity; they have poor sensitivity to lesions with blurred boundaries or small targets; and they are difficult to adapt to complex and varied lesion morphologies, exhibiting poor versatility.

[0005] With the development of deep learning technology, especially the widespread application of convolutional neural networks (CNNs) in the field of image understanding, medical image segmentation has entered a new stage dominated by end-to-end training. U-Net and its 3D extended version 3D-UNet have achieved remarkable results in tasks such as tumor segmentation and organ recognition. They can automatically learn image features, reducing reliance on prior knowledge and manual design.

[0006] However, most existing deep learning models adopt a general modeling approach, which mainly focuses on improving edge feature extraction and contextual information association. Although they have improved accuracy to some extent, they lack the ability to adapt to individual differences and ignore the differences among patients in physiological structure, disease development, tissue density, etc., which limits their application effect in precision medicine. Summary of the Invention

[0007] To address the aforementioned problems, this invention proposes a method and system for segmenting colorectal cancer CT images that integrates individualized features. This invention utilizes structured patient information (such as age, gender, family history, etc.) to guide the model network to adjust its feature response, thereby improving the model's segmentation accuracy and individual adaptability, and further enhancing the accuracy of segmentation.

[0008] According to some embodiments, the present invention adopts the following technical solution: A method for segmenting colorectal cancer CT images by incorporating individualized features, comprising the following steps: Acquire abdominal CT images of historical patients and the target patient; Obtain structured individual characteristics information related to the patient, including age, gender, family history of colorectal cancer, and history of intestinal diseases; Preprocessing of abdominal CT images; The image segmentation model was trained using preprocessed abdominal CT images of historical patients. The trained image segmentation model, which met the segmentation accuracy requirements, was then used to segment the preprocessed abdominal CT images of the target patients, generating a segmentation map of colon cancer lesions that conformed to their individual characteristics. The image segmentation model uses a 3D-UNet structure as the backbone network, with a spatial pyramid pooling module between the encoder and decoder to extract contextual semantic information at different scales. An attention gating module is added to the skipping pathway of the 3D-UNet structure to increase attention to the tumor part. It also includes an individual feature encoding subnetwork, which maps the patient's structured individual feature information into a personalized feature vector, and influences the activation state or feature channel weights of the intermediate layers of the backbone network through feature modulation, thereby realizing the individual dynamic adjustment of the segmentation strategy.

[0009] As an alternative implementation, the preprocessing of abdominal CT images includes sample enhancement through random rotation, displacement, scaling, and flipping.

[0010] As an alternative implementation, the backbone network of the image segmentation model includes an encoder and a decoder, wherein the encoder includes multiple convolutional modules, and the structure of the decoder is symmetrical to the structure of the encoding layer.

[0011] As an alternative implementation, the encoder includes four convolutional modules, each of which includes two 3×3 convolution operations and a ReLU activation function, and is downsampled using 2×2 max pooling, with the number of channels increasing layer by layer.

[0012] As an alternative implementation, the decoder includes four convolutional modules that are upsampled and concatenated with the features of the encoder. After that, the modules undergo two 3×3 convolutions and a ReLU activation function to gradually restore the spatial resolution and reduce the number of channels.

[0013] As an alternative implementation, the spatial pyramid pooling module includes multiple parallel convolutional branches with different porosity. Each parallel convolutional branch is used to extract contextual semantic features of the lesion region at different receptive field scales. The output features of each parallel convolutional branch are concatenated in the channel dimension and fused through 1×1 convolution to enhance the model's ability to perceive different morphological, size, and positional changes in the tumor region, thereby improving segmentation robustness and accuracy.

[0014] As an alternative implementation, the attention gating module is used to dynamically adjust the texture features of the encoder using the semantic information of the decoder, thereby suppressing irrelevant background interference and enhancing the response of the tumor region to improve segmentation accuracy.

[0015] As an alternative implementation, the individual feature encoding subnetwork is used to normalize or discretize the structured individual feature information, represent multiple individual variables as vectors and concatenate them, generate individual feature codes through fully connected mapping, and finally output scaling factors and offset factors by two layers of multilayer perceptrons to achieve channel-by-channel adjustment of the intermediate feature maps of the backbone network, thereby realizing the personalized segmentation strategy.

[0016] As an alternative implementation, the image segmentation model is optimized during training by using a weighted combination of the main loss function and the auxiliary loss function. The main loss function is a hybrid of Dice loss and Focal loss, while the auxiliary loss function is calculated independently for each supervised branch to improve the model's ability to recognize multi-scale lesion boundaries and fine-grained structures. The Adam optimizer is used to update the model parameters, and a learning rate warm-up mechanism and a Cosine decay strategy are introduced to control the gradient update magnitude during training, thereby improving the model's convergence efficiency and training stability.

[0017] As an alternative implementation, the evaluation metrics for segmentation accuracy include Dice similarity coefficient, IoU (Intersection over Union), sensitivity, and specificity, which are used to measure the spatial overlap, recall, and background exclusion capabilities between the model's predicted segmentation results and manually labeled data, respectively. The preprocessed abdominal CT images of historical patients are divided into training and test sets. The trained image segmentation model is validated using the test set. If the segmentation accuracy reaches or exceeds the set value, the validation is successful.

[0018] A colorectal cancer CT image segmentation system incorporating individualized features, comprising: The image acquisition module is configured to acquire abdominal CT images of historical patients and the target patient. The data acquisition module is configured to acquire structured individual characteristic information related to the patient, including age, gender, family history of colorectal cancer, and history of intestinal diseases. The preprocessing module is configured to preprocess abdominal CT images; The image segmentation module is configured to train the image segmentation model using preprocessed abdominal CT images of historical patients, and then use the trained image segmentation model that meets the segmentation accuracy requirements to segment the preprocessed abdominal CT images of the target patient to generate a segmentation map of colon cancer lesions that conforms to the individual characteristics of the patient. The image segmentation model uses a 3D-UNet structure as the backbone network, with a spatial pyramid pooling module between the encoder and decoder to extract contextual semantic information at different scales. An attention gating module is added to the skipping pathway of the 3D-UNet structure to increase attention to the tumor part. It also includes an individual feature encoding subnetwork, which maps the patient's structured individual feature information into a personalized feature vector, and influences the activation state or feature channel weights of the intermediate layers of the backbone network through feature modulation, thereby realizing the individual dynamic adjustment of the segmentation strategy.

[0019] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention achieves personalized colorectal cancer CT image segmentation by integrating individualized feature information (such as age, gender, family history of colorectal cancer, etc.), and can dynamically adjust the segmentation strategy to provide more accurate individualized diagnosis.

[0020] This invention combines a 3D-UNet structure with a Spatial Pyramid Pooling Module (ASPP) to enhance the perception of tumors at different scales, improving segmentation accuracy and robustness. The addition of an attention gating mechanism helps focus on the tumor region and suppress background noise, further improving segmentation accuracy. Furthermore, by introducing a residual connection mechanism and optimizing the training strategy, the stability and training efficiency of the deep network are enhanced, thus maintaining high segmentation performance in complex tumor regions (such as intestinal bends and densely packed tissue areas). Overall, this invention demonstrates the advantages of personalization, precision, and efficiency in colorectal cancer image segmentation.

[0021] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0022] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0023] Figure 1 This is a schematic diagram of a colon cancer CT image segmentation method in one embodiment; Figure 2 This is a backbone network architecture diagram in one embodiment; Figure 3 This is a flowchart of the PSE module in one embodiment; Figure 4 This is a flowchart of the ASPP module in one embodiment; Figure 5 This is a flowchart of an attention gating mechanism in one embodiment. Detailed Implementation

[0024] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0025] It should be noted that the following detailed description is illustrative and intended to provide further explanation of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0026] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0027] Where there is no conflict, the embodiments and features described in this application may be combined with each other.

[0028] Example 1 A segmentation method for colorectal cancer CT images that incorporates individualized features, such as Figure 1 As shown, it includes the following steps: S1. Obtain abdominal CT images of historical patients and obtain structured individual characteristic information related to the patients, including age, gender, family history of colorectal cancer, history of intestinal diseases, etc. S2. Preprocess the CT images, including slicing, rotating, normalizing, etc., to standardize the image input; S3. Construct a personalized 3D convolutional neural network model that integrates individual features, the model comprising: Its main architecture is an image segmentation backbone network based on the 3D-UNet structure; The Spatial Pyramid Pooling Module (ASPP), located between the encoder and decoder, is used to extract contextual semantic information at different scales. An attention gating module was added to the skipping pathway in 3D-UNet to increase the model's attention to the tumor region; It also has an Individual Feature Encoding Subnetwork (PSE module) to map the patient's structured features into individual feature vectors and influence the activation state or feature channel weights of the intermediate layers of the backbone network through feature modulation, thereby achieving individual dynamic adjustment of the segmentation strategy.

[0029] S4. Divide the preprocessed abdominal CT images and individual feature vectors into training set and test set. The training set needs to be labeled for the model to learn from. Provide the labeled training set and individual feature vectors to the model for training. S5. Use the trained model to conduct experiments on an unlabeled test set and verify the segmentation accuracy. If the segmentation accuracy reaches 95% or higher, it can be put into use. Otherwise, the number of training sets can be increased and the process can be returned to step S4 for retraining. S6. Use the qualified model to infer the abdominal CT images of the target patient and generate a segmentation map of colon cancer lesions that conforms to the individual characteristics of the patient.

[0030] In this embodiment, step S1 involves acquiring abdominal CT images of historical patients. These images should be high-resolution and clearly show the colon region and lesions. During CT image acquisition, the slice thickness and image quality should be adapted to the accurate localization of colon cancer lesions. In addition to CT images, structured individual characteristics of historical patients should also be acquired, including age, gender, family history of colon cancer, and intestinal disease history. All these structured individual characteristics will be converted into feature vectors through standardization and normalization, serving as auxiliary information in subsequent model training to improve the personalized adaptability of the segmentation results.

[0031] In this embodiment, step S2, patient information preprocessing, includes two parts: CT image preprocessing and patient feature information preprocessing. The purpose of CT image preprocessing is to provide data for model training. Effective preprocessing not only helps model training but also reduces the amount of training data required and prevents overfitting. In our invention, the CT image preprocessing method includes rotation, translation, and scaling. The preprocessed image provides high-quality input to the deep learning model, ensuring the effectiveness of the segmentation task. The purpose of patient feature information preprocessing is to convert information with different units of measurement, such as age and medical history, into computer language using one-hot encoding. This facilitates subsequent processing and fusion of information by the PSE module, allowing the model to understand the information. The processing formula is as follows: , .

[0032] It is assumed that each patient has Kc categorical attributes {c i} (such as gender, pathology type) and Kn numerical attributes {n j (e.g., age, BMI), onehot is a one-hot encoding operation, and normal is a normalization operation.

[0033] In this embodiment, in step S3, a personalized three-dimensional convolutional neural network model that integrates individual features is constructed. This model is based on the 3D-UNet network architecture and improves the accuracy and personalized adaptability of the model in the colorectal cancer CT image segmentation task through the integration of multiple modules.

[0034] The main structure of this model consists of the following parts: 3D-UNet Image Segmentation Backbone Network: 3D-UNet is a convolutional neural network designed for 3D medical image segmentation tasks. Its network structure diagram is shown in a symmetrical "U" shape. The specific network architecture diagram is shown below. Figure 2 The network's encoder extracts multi-scale features of the image through layer-by-layer convolution and max-pooling operations, while the decoder restores the spatial resolution of the image through upsampling and uses skip connections to combine the high-resolution features of the encoder with the low-resolution features of the decoder, ultimately generating a high-precision segmented image.

[0035] Individual Feature Encoding Subnetwork (PSE Module): Its feature flowchart is shown in Figure 3 This module maps a patient's individual structured characteristics (such as age, gender, family history, etc.) into personalized feature vectors, and influences the activation state or feature channel weights of the network's intermediate layers through feature modulation, thereby achieving a personalized segmentation strategy. This process uses a multilayer perceptron (MLP) to map patient features and outputs scaling and offset factors to adjust the network's feature map. The final output features are fused with image features, enabling the network to dynamically adjust its segmentation strategy based on the characteristics of different patients.

[0036] Spatial Pyramid Pooling Module (ASPP): Its operation flowchart is shown in Figure 4 To enhance the model's ability to perceive lesion regions at different scales, a spatial pyramid pooling module was added between the encoder and decoder. This module extracts contextual information about the tumor region across the multi-scale receptive field through convolutional branches with different dilation rates. The output features are concatenated along the channel dimension and fused using 1×1 convolutions to enhance the perception of tumor regions with varying shapes, sizes, and locations.

[0037] Attention Gating Module: Its operation flowchart is shown in Figure 5 In the skip connection path of 3D-UNet, an attention gating module is added. The key mechanism of this module is to use the semantic information at the decoding end to adjust the texture features at the encoding end, thereby improving the model's focus on the tumor region, suppressing interference from irrelevant backgrounds, and ultimately improving the accuracy of tumor segmentation.

[0038] In this embodiment, in step S4, the processed patient CT images and individual feature vectors are divided into a training set and a validation set. This dataset partitioning ensures that the model does not overfit during training and has good generalization ability in practical applications. The training set contains labeled CT images and corresponding patient individual feature information (such as age, gender, etc.). This data is used to train the model's segmentation ability. Each image in the training set has accurate tumor region annotations for supervised learning. The validation set is used to monitor the model's performance during training, ensuring the model's segmentation accuracy and generalization ability. This set does not contain any labeled data and is only used during training to evaluate the performance of the trained model.

[0039] The following methods are used to optimize the training performance of the model: Loss function design: A weighted combination of Dice loss and Focal loss is used to train the model. Dice loss measures the similarity between the predicted segmented region and the ground truth region, while Focal loss specifically addresses class imbalance, being particularly sensitive for detecting small tumor regions. The combined formula is as follows:

[0040] in, To measure the overlap of segmented regions, It is primarily used to increase attention to difficult samples (such as small tumors).

[0041] Optimizer and training strategy: The Adam optimizer is used to update the model parameters, and a learning rate warm-up mechanism and cosine decay strategy are introduced to control the gradient update magnitude during training, thereby improving the convergence speed and training stability of the model.

[0042] In this embodiment, in step S5, the trained model needs to undergo rigorous testing and validation. The model is evaluated using a test set, and the evaluation metrics include Dice similarity coefficient, IoU (Intersection over Union), sensitivity, and specificity.

[0043] Dice similarity coefficient and Intersection over Union (IoU) are commonly used image segmentation evaluation metrics. Dice measures the overlap between the predicted and ground truth regions; a higher value indicates greater similarity between the segmentation result and the ground truth annotation. IoU, calculated as the ratio of the intersection to the union of the predicted and ground truth regions, reflects the model's segmentation accuracy. Sensitivity measures the model's recall ability for tumor regions, i.e., the proportion of tumors that can be correctly identified, while specificity assesses the model's ability to correctly exclude background and reduce false positives. These metrics comprehensively reflect the model's precision, recall, and error control capabilities in the segmentation task.

[0044] The calculation formula is as follows:

[0045]

[0046]

[0047]

[0048] Here, TP indicates that a sample that is actually a positive class was correctly predicted as a positive class.

[0049] TN indicates that a sample that was actually a negative class was correctly predicted as a negative class.

[0050] FP means that a sample that is actually a negative class was incorrectly predicted as a positive class.

[0051] FN indicates that a sample that is actually a positive class was incorrectly predicted as a negative class.

[0052] These evaluations quantify the model's segmentation accuracy. If the evaluation results show a segmentation accuracy of 95% or higher, the model is considered to meet clinical application requirements and can be put into use; if the accuracy is insufficient, it can return to the training phase, increase the dataset, and continue training. In this embodiment, the score after fusing the four indicators reaches 95%, meaning the segmentation accuracy reaches 95%, and the fusion weights are set according to requirements.

[0053] In this embodiment, in step S6, the trained and validated model will be applied to inference from the CT images of new patients. The specific process is as follows: Obtain abdominal CT images of new patients and their corresponding individual feature vectors.

[0054] The preprocessed image and individual feature vectors are used as inputs and fed into the trained and validated model.

[0055] The model generates a segmentation map of colon cancer lesions that reflects the individual characteristics of each patient. This segmentation map provides doctors with precise tumor localization information, helping to develop personalized treatment plans. In this way, the present invention enables personalized and precise segmentation of colon cancer images, providing patients with higher quality medical services.

[0056] In summary, the method in this embodiment combines a spatial pyramid pooling module, an attention gating module, and an individual feature encoding subnetwork to enhance the model's ability to perceive changes in the morphology, size, and location of different tumors, and dynamically adjust the segmentation strategy according to the individual characteristics of the patient; it can introduce personalized features during the segmentation process, thereby improving the model's adaptability to different patient images.

[0057] A colorectal cancer CT image segmentation system incorporating individualized features, comprising: The image acquisition module is configured to acquire abdominal CT images of historical patients and the target patient. The data acquisition module is configured to acquire structured individual characteristic information related to the patient, including age, gender, family history of colorectal cancer, and history of intestinal diseases. The preprocessing module is configured to preprocess abdominal CT images; The image segmentation module is configured to train the image segmentation model using preprocessed abdominal CT images of historical patients, and then use the trained image segmentation model that meets the segmentation accuracy requirements to segment the preprocessed abdominal CT images of the target patient to generate a segmentation map of colon cancer lesions that conforms to the individual characteristics of the patient. The image segmentation model uses a 3D-UNet structure as the backbone network, with a spatial pyramid pooling module between the encoder and decoder to extract contextual semantic information at different scales. An attention gating module is added to the skipping pathway of the 3D-UNet structure to increase attention to the tumor part. It also includes an individual feature encoding subnetwork, which maps the patient's structured individual feature information into a personalized feature vector, and influences the activation state or feature channel weights of the intermediate layers of the backbone network through feature modulation, thereby realizing the individual dynamic adjustment of the segmentation strategy.

[0058] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of one or more computer-usable storage media (including, but not limited to, disk storage, etc.) containing computer-usable program code. CD - ROM It takes the form of a computer program product implemented on (such as optical memory, etc.).

[0059] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0060] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0061] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0062] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made by those skilled in the art without creative effort within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for segmenting colorectal cancer CT images by incorporating individualized features, characterized in that, Includes the following steps: Acquire abdominal CT images of historical patients and the target patient; Obtain structured individual characteristics information related to the patient, including age, gender, family history of colorectal cancer, and history of intestinal diseases; Preprocessing of abdominal CT images; The image segmentation model was trained using preprocessed abdominal CT images of historical patients. The trained image segmentation model, which met the segmentation accuracy requirements, was then used to segment the preprocessed abdominal CT images of the target patients, generating a segmentation map of colon cancer lesions that conformed to their individual characteristics. The image segmentation model uses a 3D-UNet structure as the backbone network, with a spatial pyramid pooling module between the encoder and decoder to extract contextual semantic information at different scales. An attention gating module is added to the skipping pathway of the 3D-UNet structure to increase attention to the tumor part. It also includes an individual feature encoding subnetwork, which maps the patient's structured individual feature information into a personalized feature vector, and influences the activation state or feature channel weights of the intermediate layers of the backbone network through feature modulation, thereby realizing the individual dynamic adjustment of the segmentation strategy.

2. The colorectal cancer CT image segmentation method incorporating individualized features as described in claim 1, characterized in that, The preprocessing of abdominal CT images includes sample enhancement through random rotation, displacement, scaling, and flipping.

3. The colorectal cancer CT image segmentation method incorporating individualized features as described in claim 1, characterized in that, The backbone network of the image segmentation model includes an encoder and a decoder, wherein the encoder includes multiple convolutional modules, and the structure of the decoder is symmetrical to the structure of the encoding layer.

4. The colorectal cancer CT image segmentation method incorporating individualized features as described in claim 1, characterized in that, The encoder includes four convolutional modules, each of which includes two 3×3 convolution operations and a ReLU activation function, and is downsampled using 2×2 max pooling, with the number of channels increasing layer by layer; The decoder includes four convolutional modules, which are concatenated with the features of the encoder through upsampling operations, and then subjected to two 3×3 convolutions and a ReLU activation function to gradually restore the spatial resolution and reduce the number of channels.

5. The colorectal cancer CT image segmentation method incorporating individualized features as described in claim 1, characterized in that, The spatial pyramid pooling module includes multiple parallel convolutional branches with different dilation rates. Each parallel convolutional branch is used to extract contextual semantic features of the lesion region at different receptive field scales. The output features of each parallel convolutional branch are concatenated in the channel dimension and fused through 1×1 convolution to enhance the model's ability to perceive different morphological, size, and positional changes in the tumor region, thereby improving segmentation robustness and accuracy.

6. The colorectal cancer CT image segmentation method incorporating individualized features as described in claim 1, characterized in that, The attention gating module is used to dynamically adjust the texture features of the encoder using the semantic information of the decoder, thereby suppressing irrelevant background interference and enhancing the response of the tumor region to improve segmentation accuracy.

7. The colorectal cancer CT image segmentation method incorporating individualized features as described in claim 1, characterized in that, The individual feature encoding subnetwork is used to normalize or discretize the structured individual feature information, represent multiple individual variables as vectors and concatenate them, generate individual feature codes through fully connected mapping, and finally output scaling factors and offset factors by two layers of multilayer perceptrons to achieve channel-by-channel adjustment of the intermediate feature maps of the backbone network, thereby realizing the personalized segmentation strategy.

8. The colorectal cancer CT image segmentation method incorporating individualized features as described in claim 1, characterized in that, During training, the image segmentation model is optimized using a weighted combination of the main loss function and the auxiliary loss function. The main loss function is a hybrid of Dice loss and Focal loss, while the auxiliary loss function is calculated independently for each supervised branch to improve the model's ability to identify multi-scale lesion boundaries and fine-grained structures. The Adam optimizer is used to update the model parameters, and a learning rate warm-up mechanism and a Cosine decay strategy are introduced to control the gradient update magnitude during training, thereby improving the model's convergence efficiency and training stability.

9. The colorectal cancer CT image segmentation method incorporating individualized features as described in claim 1, characterized in that, The evaluation metrics for segmentation accuracy include Dice similarity coefficient, IoU (Intersection over Union), sensitivity, and specificity, which are used to measure the spatial overlap, recall, and background exclusion capabilities between the model's predicted segmentation results and manually labeled data, respectively. The preprocessed abdominal CT images of historical patients are divided into training and test sets. The trained image segmentation model is validated using the test set. If the segmentation accuracy reaches or exceeds the set value, the validation is successful.

10. A colorectal cancer CT image segmentation system incorporating individualized features, characterized in that, include: The image acquisition module is configured to acquire abdominal CT images of historical patients and the target patient. The data acquisition module is configured to acquire structured individual characteristic information related to the patient, including age, gender, family history of colorectal cancer, and history of intestinal diseases. The preprocessing module is configured to preprocess abdominal CT images; The image segmentation module is configured to train the image segmentation model using preprocessed abdominal CT images of historical patients, and then use the trained image segmentation model that meets the segmentation accuracy requirements to segment the preprocessed abdominal CT images of the target patient to generate a segmentation map of colon cancer lesions that conforms to the individual characteristics of the patient. The image segmentation model uses a 3D-UNet structure as the backbone network, with a spatial pyramid pooling module between the encoder and decoder to extract contextual semantic information at different scales. An attention gating module is added to the skipping pathway of the 3D-UNet structure to increase attention to the tumor part. It also includes an individual feature encoding subnetwork, which maps the patient's structured individual feature information into a personalized feature vector, and influences the activation state or feature channel weights of the intermediate layers of the backbone network through feature modulation, thereby realizing the individual dynamic adjustment of the segmentation strategy.

Images