AI-based automated liver cancer tissue recognition and diagnosis system

Abstract

The present invention discloses an AI-based automated liver cancer tissue recognition and diagnosis system, which relates to the field of medical impact processing technology, including a data upload terminal, an image processing terminal, a liver cancer tissue algorithm recognition terminal, a liver cancer automatic diagnosis terminal and a user interface terminal; the data upload terminal is responsible for uploading the patient's liver image data to the cloud or local analysis platform of the system through a secure transmission channel for processing; the image processing terminal is responsible for preprocessing the acquired image data, including denoising, contrast enhancement and artifact correction operations; using a filtering algorithm to remove noise in the image; and using adaptive histogram equalization to increase the image contrast. The present invention can automatically extract the subtle features of liver cancer lesions and improve the recognition accuracy of different lesion morphologies; compared with the traditional subjective judgment that relies on the doctor's experience, the system shows higher sensitivity in detecting small lesions and early liver cancer, reducing the probability of misdiagnosis and missed diagnosis.

Description

Technical Field

[0001] The present invention relates to the field of medical image processing technology, and in particular to an AI-based automated liver cancer tissue recognition and diagnosis system. Background Art

[0002] Liver cancer is one of the most common malignant tumors with the highest morbidity and mortality worldwide. Early diagnosis is of great significance to improving the survival rate of patients. Traditional diagnosis of liver cancer mainly relies on imaging examinations such as ultrasound, computed tomography and magnetic resonance imaging, combined with pathological analysis to determine the nature of the lesion.

[0003] However, the current imaging diagnosis process is highly dependent on the doctor's experience and has the following technical defects: for early liver cancer, the imaging features are often not obvious, and the existing technology has certain limitations in the accuracy of automatic detection of liver cancer lesions, especially in the case of small lesions or blurred tumor boundaries, which are prone to missed diagnosis and misdiagnosis. In addition, the existing computer-aided diagnosis system can often only perform basic image enhancement or edge detection, and cannot effectively extract the deep features of the tumor for identification, so there is room for improvement. Summary of the invention

[0004] The purpose of this invention is to solve the shortcomings of the prior art and propose an AI-based automated liver cancer tissue recognition and diagnosis system. Its advantage is that it can automatically extract the subtle features of liver cancer lesions, improving the recognition accuracy of different lesion morphologies; compared with the traditional subjective judgment that relies on the doctor's experience, the system shows higher sensitivity in detecting small lesions and early liver cancer, reducing the probability of misdiagnosis and missed diagnosis.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] AI-based automated liver cancer tissue recognition and diagnosis system, including data upload end, image processing end, liver cancer tissue algorithm recognition end, liver cancer automatic diagnosis end and user interface end;

[0007] The data upload end is responsible for uploading the patient's liver image data to the system's cloud or local analysis platform through a secure transmission channel for processing; the image processing end is responsible for preprocessing the acquired image data, including denoising, contrast enhancement and artifact correction operations; using a filtering algorithm to remove noise in the image; and using adaptive histogram equalization to increase the image contrast;

[0008] The liver cancer tissue algorithm recognition end is responsible for using the deep learning model to analyze the uploaded image data using AI image recognition technology to automatically identify the presence of liver cancer tissue and its specific area; after the model analysis is completed, the system automatically generates a boundary map of the tumor tissue, lesion location and size information, and marks the suspected cancerous area;

[0009] The liver cancer automatic diagnosis terminal is responsible for automatically generating a preliminary diagnosis result of liver cancer based on tumor tissue recognition, combined with clinical data and imaging features, and providing corresponding treatment suggestions; it assists doctors in completing the diagnosis by comprehensively analyzing the size, location and spread of the tumor;

[0010] The user interface is responsible for displaying AI analysis results and diagnostic reports, and allows medical experts or operators to perform further interactive operations; it intuitively displays the analysis results of the image, including the marked tumor area, liver segmentation results, tumor size and location, and supports doctors to zoom in, zoom out or rotate the tumor lesions for viewing.

[0011] The present invention is further configured such that the filtering algorithm includes a Gaussian filtering algorithm and a mean filtering algorithm.

[0012] The present invention is further configured such that the liver cancer tissue algorithm recognition end uses a deep learning model based on a convolutional neural network and a Transformer architecture. The model is trained with a large number of annotated data sets and can accurately segment the liver and tumor areas.

[0013] The present invention is further configured such that the automatic liver cancer diagnosis terminal analyzes and identifies tumor features, such as shape, irregularity, texture and boundaries, and combines the patient's medical history data (such as age, gender and history of cirrhosis) so that AI can infer the type of tumor and its stage.

[0014] The present invention is further configured such that the automatic liver cancer diagnosis end relies on a pre-trained model to predict the staging and classification of liver cancer, determines whether it is primary liver cancer or metastatic liver cancer, and gives a diagnosis based on the TNM (tumor, lymph node and metastasis) staging system; based on the diagnosis results, automatically generates corresponding treatment recommendations, including surgery, radiotherapy, chemotherapy or interventional treatment, and provides a reference plan in combination with medical guidelines.

[0015] The present invention is further configured such that the user interface supports users to review images by mouse clicks or touch operations, doctors can manually annotate or correct AI segmentation results, and allow new image data to be imported for real-time analysis; access is made through a Web browser or an independent client, supporting doctors to remotely view and operate, allowing doctors to access the system at different locations.

[0016] The present invention is further configured such that the convolutional neural network algorithm includes data collection, data labeling, and data segmentation and preparation, wherein the data collection mainly comes from imaging data of liver cancer patients from cooperative hospitals, and the imaging types include CT and MRI; the internal tissue structure of the liver, especially the changes in tumor location and shape, are captured to cover a variety of situations and a variety of patient groups; including patients of different ages, genders, and degrees of lesions.

[0017] The present invention is further configured such that the data annotation uses a professional medical image annotation tool labelme to perform detailed annotations according to the tumor position and morphology of each image; the doctor outlines the edge of the tumor using a fine polygon tool based on the image characteristics; according to the characteristics and classification of the medical image data, the liver and tumor areas need to be classified during annotation; after the initial annotation is completed, all annotation results need to be proofread by experts for a second time, and for more complex or difficult cases, a double-blind annotation method is used, with two doctors annotating separately and finally unifying the results.

[0018] The present invention is further configured such that the data segmentation and preparation are used to divide the collected image data into a training set (70%), a validation set (15%) and a test set (15%) according to a certain ratio; the training set is used for learning the model, the validation set is used for adjusting parameters in the model optimization process, and the test set is used for finally evaluating the performance of the model; since the proportion of tumor tissue in the liver is usually small, the number of samples in the tumor area is much lower than that in the liver and background areas, so the data needs to be balanced through resampling, data enhancement and generative adversarial network methods.

[0019] The present invention is further configured as follows: the convolutional neural network algorithm often adopts a loss function based on cross entropy loss, Dice coefficient and IoU index during training to measure the gap between the segmentation result and the true label; the target in liver cancer segmentation is relatively sparse, and the Dice loss function is introduced to balance the imbalance of positive and negative samples; the convolutional neural network algorithm model is trained by a medical imaging data set, the performance of the model is evaluated in real time using a validation set, and the model is continuously optimized by adjusting the learning rate, batch size and hyperparameters; in actual applications, the model accepts new input data and performs inference in real time to generate segmentation results of the liver and tumor for doctors' reference.

[0020] The beneficial effects of the present invention are:

[0021] 1. The present invention uses advanced AI algorithms, especially image recognition models based on convolutional neural networks and deep learning, to automatically extract subtle features of liver cancer lesions, thereby improving the recognition accuracy of different lesion morphologies (such as blurred tumor boundaries and irregular lesions). Compared with traditional subjective judgments that rely on doctors' experience, the system shows higher sensitivity in detecting small lesions and early liver cancer, reducing the probability of misdiagnosis and missed diagnosis.

[0022] 2. Traditional liver cancer diagnosis often requires doctors to manually analyze a large number of images, which is time-consuming and labor-intensive, and is particularly inefficient when dealing with complex cases. The present invention uses automated image processing and rapid analysis technology to complete the recognition and analysis of large-scale medical image data in a short period of time, thereby significantly shortening the diagnosis time, improving the work efficiency of medical institutions, and reducing patient waiting time.

[0023] 3. The present invention not only supports the analysis of a single imaging mode, but can also integrate multimodal imaging data such as CT, MRI, and ultrasound for comprehensive diagnosis, providing more comprehensive tumor tissue information; this multimodal image fusion technology can help doctors more accurately evaluate the size, location, and spread of tumors from different angles, thereby providing more comprehensive auxiliary information for clinical treatment.

[0024] 4. The present invention can identify the lesion area in real time during liver cancer surgery, provide surgical path planning and boundary resection suggestions, and optimize the surgical plan; compared with traditional surgery that relies on the doctor's intraoperative experience and judgment, the system can reduce the risk of incomplete tumor resection or intraoperative bleeding, improve the safety and accuracy of the surgery, and reduce postoperative complications.

[0025] 5. The present invention performs automated analysis based on AI algorithms, and the diagnostic results are highly consistent and standardized. Unlike traditional diagnosis that relies on individual doctor experience, the system can eliminate diagnostic bias caused by experience differences, ensure consistency in diagnostic quality between different patients and doctors, and improve the reliability of diagnosis.

[0026] 6. The present invention is not only an automated tool, but also can serve as an auxiliary decision-making system for doctors. It can reduce the workload of doctors by automatically marking liver cancer lesions, analyzing the spread of tumors, and providing treatment suggestions. In particular, it can help doctors quickly focus on key areas when faced with large amounts of imaging data, thereby effectively supporting clinical decision-making.

[0027] 7. The present invention can provide accurate liver cancer lesion identification and diagnosis results, which can provide strong support for subsequent treatment plans, help doctors develop more personalized and accurate treatment plans, improve patients' prognosis, and improve the overall treatment quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 This is a schematic diagram of the overall structure of the AI-based automated liver cancer tissue recognition and diagnosis system proposed by the present invention;

[0029] Figure 2 This is a schematic diagram of the image processing module of the AI-based automated liver cancer tissue recognition and diagnosis system proposed by the present invention;

[0030] Figure 3A block diagram of the liver cancer tissue recognition algorithm of the AI-based liver cancer tissue automatic recognition and diagnosis system proposed by the present invention;

[0031] Figure 4 This is a schematic diagram of the training process of the AI-based automatic liver cancer tissue recognition and diagnosis system proposed by the present invention;

[0032] Figure 5 This is a schematic diagram of the model loading and use process of the AI-based liver cancer tissue automatic identification and diagnosis system proposed in the present invention. DETAILED DESCRIPTION

[0033] The technical solution of this patent is further described in detail below in conjunction with specific implementation methods.

[0034] The embodiments of the present invention are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be understood as limiting the present invention.

[0035] Reference Figure 1-3 , an AI-based liver cancer tissue automatic identification and diagnosis system, including a data upload terminal, an image processing terminal, a liver cancer tissue algorithm identification terminal, a liver cancer automatic diagnosis terminal, and a user interface terminal;

[0036] The data upload end is responsible for uploading the patient's liver imaging data to the system's cloud or local analysis platform for processing through a secure transmission channel; it supports common medical imaging formats, such as JPEG, PNG, etc., and the system can process imaging data from multiple sources. The uploaded imaging data undergoes preliminary preprocessing steps, including image standardization, denoising, and format conversion, to ensure that images from different sources can be processed uniformly. The upload end uses encrypted transmission protocols (HTTPS, other medical-grade encryption protocols) to protect patient privacy and comply with medical data security standards such as HIPAA.

[0037] The image processing end is responsible for preprocessing the acquired image data, including denoising, contrast enhancement and artifact correction operations; using filtering algorithms to remove noise in the image; using adaptive histogram equalization to increase the contrast of the image and enhance the clarity of the image; filtering algorithms include Gaussian filtering algorithm and mean filtering algorithm;

[0038] The liver cancer tissue algorithm recognition end is responsible for using the deep learning model to analyze the uploaded image data using AI image recognition technology, automatically identifying the presence of liver cancer tissue and its specific area; it can quickly and accurately segment the liver area and detect tumor lesions; after the model analysis is completed, the system automatically generates a boundary map of the tumor tissue, lesion location and size information, and marks the suspected cancerous area; using a deep learning model based on a convolutional neural network (U-Net algorithm) and Transformer architecture, the model has been trained with a large number of labeled data sets and can accurately segment the liver and tumor areas;

[0039] The automatic liver cancer diagnosis terminal is responsible for automatically generating preliminary diagnosis results for liver cancer based on tumor tissue identification, combined with clinical data and imaging features, and providing corresponding treatment recommendations; it assists doctors in completing the diagnosis by comprehensively analyzing the size, location and spread of the tumor; by analyzing the identified tumor features, such as shape, irregularity, texture and boundaries, combined with the patient's medical history data (such as age, gender and history of cirrhosis), AI can infer the type of tumor and its stage; it relies on pre-trained models to predict the stage and classification of liver cancer, determine whether it is primary liver cancer or metastatic liver cancer, and give a diagnosis based on the TNM (tumor, lymph node and metastasis) staging system; based on the diagnosis results, it automatically generates corresponding treatment recommendations, including surgery, radiotherapy, chemotherapy or interventional treatment, and provides reference plans in combination with medical guidelines;

[0040] The user interface is responsible for displaying AI analysis results and diagnostic reports, and allows medical experts or operators to perform further interactive operations; it intuitively displays the analysis results of the image, including the marked tumor area, liver segmentation results, tumor size and location, and supports doctors to zoom in, reduce or rotate the tumor lesions; it supports users to review images by clicking the mouse or touching the screen, and doctors can manually mark or correct AI segmentation results, while allowing new image data to be imported for real-time analysis; it supports remote viewing and operation by doctors through a Web browser or an independent client, allowing doctors to access the system from different locations to improve work efficiency.

[0041] Convolutional neural network algorithms include data collection, data labeling, and data segmentation and preparation;

[0042] 1. Data Collection

[0043] 1.1. Image source

[0044] Multi-center medical imaging data: The system's data mainly comes from imaging data of liver cancer patients in cooperative hospitals, including but not limited to CT (Computed Tomography), MRI (Magnetic Resonance Imaging), etc. These images capture the internal tissue structure of the liver, especially the changes in the location and shape of the tumor.

[0045] CT images: can clearly show the volume and boundaries of the liver and tumors.

[0046] MRI images: They have a higher resolution for soft tissue and can provide detailed information about the internal tissues of the tumor.

[0047] 1.2. Data size

[0048] In order to improve the generalization ability of the model, the data should cover a variety of situations and patient groups as much as possible, including patients of different ages, genders, and lesion severity, to ensure that the model can have a high recognition accuracy rate for different individual lesion conditions in actual applications.

[0049] Data scale: Collect liver cancer imaging data from 500 patients to ensure that the model can recognize diverse tumor morphologies, locations, and sizes.

[0050] 1.3. Data normalization and desensitization

[0051] Medical imaging data usually contains patients' personal information. Therefore, during the data collection stage, data privacy protection regulations (such as the Personal Information Protection Law and GDPR) must be strictly observed, and all imaging data must be de-identified (desensitized) to ensure that privacy is not leaked.

[0052] Image data also needs to be standardized, such as converting images of different formats into a unified DICOM format or other commonly used standard medical image formats to facilitate subsequent processing and model input.

[0053] 2. Data Annotation

[0054] 2.1. Annotation tools and processes

[0055] Labeling tools: Labeling uses the professional medical image labeling tool labelme. These tools support 3D display and labeling of medical images, and can help doctors efficiently label tumor areas and liver boundaries.

[0056] Labelers: To ensure the accuracy of data labeling, the labeling task should be completed by experienced medical experts (radiologists, oncologists). Since the boundaries of the liver and its tumors may be very vague, especially small lesion areas are particularly difficult to label, so the participation of experts is necessary.

[0057] 2.2. Annotation content

[0058] Liver region annotation: In the image data, the overall outline of the liver needs to be accurately annotated first. This is the basis for subsequent tumor identification and ensures that the model can accurately identify the boundaries of the liver.

[0059] Tumor area annotation: Detailed annotation is performed based on the location and morphology of the tumor in each image. Based on the image characteristics, the doctor outlines the edge of the tumor with a fine polygon tool to ensure that the true boundary of the tumor is accurately represented.

[0060] Category labeling: According to the characteristics and classification of medical imaging data, the liver and tumor areas need to be classified during labeling. Usually, the labeling includes the following categories:

[0061] Background: Parts that are not part of the liver or tumor.

[0062] Liver: Normal liver tissue.

[0063] Tumor: Cancerous tissue in the liver.

[0064] Label standardization: To ensure data consistency, the labeling process needs to follow unified labeling specifications and standards. These standards are labeling rules developed based on international medical imaging standards, such as the TNM staging system for liver cancer.

[0065] 2.3. Data verification and proofreading

[0066] Expert proofreading: After the initial annotation is completed, all annotation results must be proofread by experts for a second time to ensure the accuracy of the annotated area. For more complex or difficult cases, a double-blind annotation method can be used, with two doctors annotating separately and finally unifying the results to ensure data accuracy.

[0067] Consistency check: The labeling results are checked for consistency by calculating indicators such as the Kappa coefficient or the Dice coefficient to ensure that the deviation between different labelers is small, especially for the labeling of the tumor edge area.

[0068] 3. Data segmentation and preparation

[0069] Training, validation and test data division: In order to ensure the generalization ability of the model, the collected image data needs to be divided into a training set (70%), a validation set (15%) and a test set (15%) according to a certain ratio. The training set is used for model learning, the validation set is used for parameter adjustment during model optimization, and the test set is used to finally evaluate the performance of the model.

[0070] Data balance: Since the proportion of tumor tissue in the liver is usually small, the number of samples in the tumor area is much lower than that in the liver and background areas. Therefore, the data needs to be balanced through methods such as resampling, data augmentation, or generative adversarial networks (GANs) to avoid overfitting of the model to the background area.

[0071] Reference Figure 4-5 ,The ,integrated neural network algorithm process includes data preprocessing, U-Net model structure, segmentation output, post-processing and model optimization;

[0072] Data preprocessing:

[0073] Input image acquisition: Obtain medical image data from the data upload end of the system. Common image formats include CT, MRI, etc. Since medical images usually have large resolution and noise, the data needs to be preprocessed to meet the input requirements of the model.

[0074] Image standardization: Normalize or standardize the input image so that the pixel value range is within a certain range (such as 0 to 1). This ensures that the input data distribution of the model is consistent, improving training results and prediction stability.

[0075] In order to enhance the generalization ability of the model, random data augmentation operations may be required on the input image data during the training phase, such as rotation, scaling, flipping, contrast adjustment, etc. This step creates more samples without changing the liver tissue and its tumor structure, thus improving the robustness of the model.

[0076] 2. U-Net model structure

[0077] U-Net is a classic convolutional neural network (CNN) architecture. Its structure is mainly divided into two parts: encoder (Contracting Path) and decoder (Expanding Path), which are connected by skip connections.

[0078] Encoder part: The encoder uses convolution operations to gradually reduce the spatial resolution of the input image and extract the features of the image.

[0079] The main process includes continuous convolution and pooling layer operations. In each convolution operation, multiple convolution kernels are applied to the input for feature extraction, and nonlinearity is introduced through activation functions (such as ReLU). The subsequent pooling layer (usually maximum pooling) is used for downsampling, halving the resolution of the feature map while retaining important spatial features.

[0080] The encoder part extracts deep features of liver and tumor tissues.

[0081] Decoder part: The decoder gradually restores the spatial resolution of the image through deconvolution or upsampling operations.

[0082] The main process includes deconvolution (or upsampling), convolution and other operations. Through the deconvolution operation, the resolution of the feature map is gradually restored to the size of the input image. After each upsampling operation, a convolution layer is added to further refine the features.

[0083] The decoder part gradually restores the spatial information of the image and reconstructs the liver and tumor areas.

[0084] Skip connections: Skip connections connect the high-resolution feature maps of the corresponding layers in the encoder with the feature maps in the decoder. In this way, the decoder can use the local detail information in the encoder to improve the model's segmentation effect on boundaries and details. This is especially important for accurate identification of liver edges and tumor tissue boundaries.

[0085] 3. Segmentation output

[0086] Prediction results: After being processed by the encoder and decoder, the output layer of U-Net is usually a fully convolutional layer, which generates a segmentation mask of the same size as the input image. Each pixel value of the mask represents the probability that the location belongs to a certain category (such as liver, tumor or background).

[0087] Category assignment: By setting a threshold (such as 0.5), the segmentation mask is binarized or multi-classified to generate the final segmentation result. Specifically, if the pixel value is greater than the threshold, the pixel is classified as liver or tumor area, otherwise it is background.

[0088] Tumor region detection: In the final segmentation results, the liver region and the tumor region are clearly segmented, and the model can mark the specific location, shape, and size of the liver edge and the tumor.

[0089] 4. Post-processing

[0090] Artifact removal: Some artifacts or noise may appear in the segmentation results. Morphological operations (opening and closing operations) can be used to remove isolated small areas, thereby improving the overall accuracy of the segmentation.

[0091] Region labeling and statistics: Label the detected tumor region, including the volume, location and number of pixels occupied by the tumor. This information will be used in subsequent diagnosis modules.

[0092] Multimodal fusion: If the system input is multimodal medical imaging data, it may be necessary to combine the segmentation results of CT and MRI to provide more accurate information for the final diagnosis of the liver and tumor.

[0093] 5. Model Optimization

[0094] Loss function: During the training process, the U-Net model often uses loss functions based on indicators such as cross entropy loss, Dice coefficient or IoU (Intersection over Union) to measure the gap between the segmentation result and the true label. The targets in liver cancer segmentation are relatively sparse, so it is suitable to introduce the Dice loss function to balance the imbalance between positive and negative samples.

[0095] Model training and validation: The U-Net model is trained using medical imaging datasets, the validation set is used to evaluate the model’s performance in real time, and the model is continuously optimized by adjusting hyperparameters such as learning rate and batch size.

[0096] Model reasoning: In actual applications, the model accepts new input data and performs reasoning in real time to generate liver and tumor segmentation results for doctors’ reference.

[0097] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. The AI-based automated liver cancer tissue identification and diagnosis system is characterized by: It includes data upload terminal, image processing terminal, liver cancer tissue algorithm recognition terminal, liver cancer automatic diagnosis terminal and user interface terminal; The data upload end is responsible for uploading the patient's liver image data to the system's cloud or local analysis platform through a secure transmission channel for processing; the image processing end is responsible for preprocessing the acquired image data, including denoising, contrast enhancement, and artifact correction operations; Use filtering algorithms to remove noise from images; use adaptive histogram equalization to increase image contrast; The liver cancer tissue algorithm recognition end is responsible for using the deep learning model to analyze the uploaded image data using AI image recognition technology to automatically identify the presence of liver cancer tissue and its specific area; after the model analysis is completed, the system automatically generates a boundary map of the tumor tissue, lesion location and size information, and marks the suspected cancerous area; The liver cancer automatic diagnosis terminal is responsible for automatically generating a preliminary diagnosis result of liver cancer based on tumor tissue recognition, combined with clinical data and imaging features, and providing corresponding treatment suggestions; it assists doctors in completing the diagnosis by comprehensively analyzing the size, location and spread of the tumor; The user interface is responsible for displaying AI analysis results and diagnostic reports, and allows medical experts or operators to perform further interactive operations; it intuitively displays the analysis results of the image, including the marked tumor area, liver segmentation results, tumor size and location, and supports doctors to zoom in, zoom out or rotate the tumor lesions for viewing.

2. The AI-based automated liver cancer tissue identification and diagnosis system according to claim 1, characterized in that: The filtering algorithms include Gaussian filtering algorithm and mean filtering algorithm.

3. The AI-based automated liver cancer tissue identification and diagnosis system according to claim 1, characterized in that: The liver cancer tissue algorithm recognition end uses a deep learning model based on convolutional neural network and Transformer architecture. The model has been trained with a large number of labeled data sets and can accurately segment the liver and tumor areas.

4. The AI-based automated liver cancer tissue identification and diagnosis system according to claim 1, characterized in that: The automatic liver cancer diagnosis terminal analyzes the identified tumor features, such as shape, irregularity, texture and boundaries, and combines the patient's medical history data (such as age, gender and history of cirrhosis). AI can infer the type of tumor and its stage.

5. The AI-based automated liver cancer tissue identification and diagnosis system according to claim 4, characterized in that: The automatic liver cancer diagnosis terminal relies on a pre-trained model to predict the staging and classification of liver cancer, determine whether it is primary liver cancer or metastatic liver cancer, and make a diagnosis based on the TNM (tumor, lymph node and metastasis) staging system; based on the diagnosis results, it automatically generates corresponding treatment recommendations, including surgery, radiotherapy, chemotherapy or interventional treatment, and provides a reference plan in combination with medical guidelines.

6. The AI-based automated liver cancer tissue identification and diagnosis system according to claim 1, characterized in that: The user interface supports users to review images by mouse clicks or touch operations. Doctors can manually annotate or correct AI segmentation results, and allow new image data to be imported for real-time analysis. Access through a web browser or an independent client supports remote viewing and operation by doctors, allowing doctors to access the system from different locations.

7. The AI-based automated liver cancer tissue identification and diagnosis system according to claim 3, characterized in that: The convolutional neural network algorithm includes data collection, data labeling, and data segmentation and preparation. The data collection mainly comes from imaging data of liver cancer patients in cooperative hospitals. The imaging types include CT and MRI. It captures the internal tissue structure of the liver, especially the changes in tumor location and shape, covering a variety of situations and patient groups, including patients of different ages, genders, and different degrees of lesions.

8. The AI-based automated liver cancer tissue identification and diagnosis system according to claim 7, characterized in that: The data were labeled using a professional medical image labeling tool, labelme, which performed detailed labeling based on the tumor location and morphology of each image. The doctor outlined the edge of the tumor using a fine polygon tool based on the image characteristics. Based on the characteristics and classification of medical image data, the liver and tumor areas needed to be classified during labeling. After the initial labeling was completed, all labeling results needed to be proofread by experts for a second time. For more complex or difficult cases, a double-blind labeling method was used, with two doctors labeling each case separately and finally unifying the results.

9. The AI-based automated liver cancer tissue identification and diagnosis system according to claim 8, characterized in that: The data segmentation and preparation are used to divide the collected imaging data into a training set (70%), a validation set (15%) and a test set (15%) according to a certain ratio; the training set is used for model learning, the validation set is used for parameter adjustment in the process of model optimization, and the test set is used for the final evaluation of the performance of the model; since the proportion of tumor tissue in the liver is usually small, the number of samples in the tumor area is much lower than that in the liver and background areas, so the data needs to be balanced through resampling, data enhancement and generative adversarial network methods.

10. The AI-based automated liver cancer tissue identification and diagnosis system according to claim 9, characterized in that: The convolutional neural network algorithm often uses a loss function based on cross entropy loss, Dice coefficient and IoU indicator during training to measure the gap between the segmentation result and the true label; the targets in liver cancer segmentation are relatively sparse, and the Dice loss function is introduced to balance the imbalance of positive and negative samples; the convolutional neural network algorithm model is trained with a medical imaging dataset, the performance of the model is evaluated in real time using a validation set, and the model is continuously optimized by adjusting the learning rate, batch size and hyperparameters; in actual applications, the model accepts new input data and performs inference in real time to generate liver and tumor segmentation results for doctors' reference.

Images