Medical image intelligent analysis system based on artificial intelligence

Abstract

The invention, which relates to the technical field of medical image intelligent analysis, discloses an artificial intelligence-based medical image intelligent analysis system comprising a data acquisition module, a data processing module, a multi-modal fusion module, a model training module, a diagnosis module and a visualization module. The data acquisition module is used for acquiring medical images from various medical imaging devices and obtaining original medical image data; the data processing module is used for preprocessing the original medical image data by adopting data cleaning, normalization and enhancement technologies to obtain a high-quality image data set; the multi-modal fusion module is used for integrating the high-quality image data set by adopting a data fusion method, and combining data from different types of imaging equipment together to generate a comprehensive feature set; and the model training module is used for training the comprehensive feature set by adopting a deep learning algorithm, and improving the model performance by adjusting model parameters and optimizing a loss function.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of medical image intelligent analysis, and particularly relates to a medical image intelligent analysis system based on artificial intelligence. BACKGROUND

[0002] Medical image intelligent analysis technology refers to a technology of using artificial intelligence (AI), especially machine learning and deep learning technology, to automatically analyze medical images obtained from various medical imaging devices. This technology aims to improve diagnostic accuracy, speed up the diagnosis process, reduce human errors, and provide decision support for doctors.

[0003] In the field of medical image intelligent analysis, medical images come from various imaging devices (such as CT, MRI, etc.), and the data formats and standards of these devices are not unified, leading to difficulties in data integration. Moreover, the information provided by a single imaging mode is limited, making it difficult to comprehensively evaluate complex pathological conditions. At the same time, spatial alignment of different modal data is also a challenge. Meanwhile, existing AI systems often lack sufficient clinical guidance and support functions, and cannot effectively assist doctors in making accurate diagnoses. SUMMARY

[0004] In view of the above existing problems, the present application is proposed.

[0005] Therefore, the present application provides a medical image intelligent analysis system based on artificial intelligence to solve the problem that medical images come from various imaging devices (such as CT, MRI, etc.), and the data formats and standards of these devices are not unified, leading to difficulties in data integration. Moreover, the information provided by a single imaging mode is limited, making it difficult to comprehensively evaluate complex pathological conditions.

[0006] To solve the above technical problems, the present application provides the following technical solutions:

[0007] In a first aspect, the present application provides a medical image intelligent analysis system based on artificial intelligence, which includes:

[0008] a data acquisition module, a data processing module, a multi-modal fusion module, a model training module, a diagnosis module, and a visualization module;

[0009] The data acquisition module is used to acquire medical images from various medical imaging devices and obtain raw medical image data.

[0010] The data processing module is used to preprocess the raw medical image data using data cleaning, normalization, and enhancement techniques to obtain a high-quality image data set.

[0011] The multi-modal fusion module is configured to integrate high-quality image datasets by using a data fusion method, combine data from different types of imaging devices together, and generate a comprehensive feature set.

[0012] The model training module is configured to train the comprehensive feature set by using a deep learning algorithm, adjust model parameters and optimize a loss function to improve model performance, and obtain a disease detection model.

[0013] The diagnosis module is configured to analyze new medical images by using the disease detection model, identify and locate possible lesion areas, and obtain a preliminary diagnosis result.

[0014] The visualization module is configured to analyze the preliminary diagnosis result obtained by the diagnosis module by using an attention mechanism or other explanatory method, and obtain a visual report.

[0015] As a preferred scheme of the medical image intelligent analysis system based on artificial intelligence, the medical images are collected from multiple medical imaging devices, and the original medical image data is obtained by the following steps:

[0016] The required medical images are extracted by connecting the hospital information system or the picture archiving and communication system through the standardized interface DICOM protocol.

[0017] Each image is attached with metadata information when collected to ensure the integrity and traceability of the data.

[0018] The metadata information includes patient ID, collection time, and device model.

[0019] For images that do not pass the quality evaluation, the contrast stretching technique is used to improve the gray dynamic range of the images, and the original medical image dataset is obtained.

[0020] As a preferred scheme of the medical image intelligent analysis system based on artificial intelligence, the original medical image data is preprocessed by using data cleaning, normalization, and enhancement techniques to obtain a high-quality image dataset by the following steps:

[0021] The metadata information of each image is checked, and the repeated or incomplete image data is deleted.

[0022] For images containing obvious errors or not meeting the collection standards, they are marked and removed.

[0023] The image files that may be damaged are detected and repaired by using the check algorithm MD5.

[0024] All images are converted to grayscale, and the intensity value of each pixel is made to fall within a specific range through a linear transformation method to eliminate differences between different imaging devices;

[0025] Image data augmentation;

[0026] The data augmentation includes rotation, flipping, and scaling operations;

[0027] During the training phase, an appropriate amount of artificial noise is added to the image, and a filter is used to remove the actual noise.

[0028] After the above steps, all images are saved in a uniform format, forming a high-quality image dataset.

[0029] As a preferred embodiment of the AI-based intelligent medical image analysis system of the present invention, the step of integrating high-quality image datasets using a data fusion method to combine data from different types of imaging devices to generate a comprehensive feature set includes the following steps:

[0030] For each type of medical image, a convolutional neural network (CNN) is used to extract key features from the image;

[0031] Rigid body transformation is used to match the same anatomical location in images of different modalities. The expression is:

[0032] T(x,y) = A·x + B·y + C;

[0033] Where T is the spatial transformation function, and A, B and C are transformation matrix parameters;

[0034] After spatial registration is completed, a weighted fusion strategy is used to combine feature vectors from different modalities to generate a comprehensive feature set, expressed as:

[0035] F fused =α·F CT +(1-α)·F MRI ;

[0036] Among them, F fused F is the fused feature vector. CT and F MRI are the feature vectors of CT and MRI, respectively, and α is the weighting parameter;

[0037] The feature vectors, after feature extraction, spatial registration, and feature fusion, are stored to form a comprehensive feature set.

[0038] As a preferred scheme of the medical image intelligent analysis system based on artificial intelligence, the comprehensive feature set is trained by using a deep learning algorithm, and model parameters are adjusted and a loss function is optimized to improve model performance to obtain a disease detection model, and the specific steps are as follows:

[0039] ResNet50 is selected as a basic model;

[0040] The comprehensive feature set is divided into a training set, a validation set and a test set;

[0041] A cross-entropy loss function is used as a target function L for training, and the expression is as follows:

[0042]

[0043] Wherein, L is a loss value, y i is a real label, is a predicted value, and N is a sample number;

[0044] The comprehensive feature vector in the training set is input into the initialized model for training, and the loss function defined above is minimized, and the expression is as follows:

[0045]

[0046] Wherein, θ is a model parameter, η is a learning rate, and t represents an iteration number;

[0047] The early stopping strategy ES is applied, and when the performance on the validation set no longer improves, the training is stopped to avoid overfitting the training data;

[0048] When the model reaches the optimal performance on the validation set, the model is comprehensively evaluated on an independent test set to ensure that the model has good generalization ability;

[0049] After the evaluation is completed, the model is deployed to a production environment for actual disease detection work.

[0050] As a preferred scheme of the medical image intelligent analysis system based on artificial intelligence, the new medical image is analyzed by using the disease detection model, and a possible lesion area is identified and located to obtain a preliminary diagnosis result, and the specific steps are as follows:

[0051] The new medical image to be analyzed is imported into the system through a DICOM standard interface;

[0052] The same preprocessing steps as in the training phase are performed on the newly input image, including size adjustment, gray scale standardization and necessary data enhancement operations;

[0053] extract feature vectors from the preprocessed new image using the same deep learning model architecture as in the training process;

[0054] When the new image comes from multiple imaging devices, then the feature vectors of these different modalities need to be fused;

[0055] A weighted fusion strategy is adopted to generate a comprehensive feature vector, expressed as:

[0056] F fused =α·F CT +(1-α)·F MRI ;

[0057] Where F fused is the fused feature vector, F CT and F MRI are the feature vectors of CT and MRI respectively, and alpha is the weight parameter, which is dynamically adjusted according to the importance of the modalities;

[0058] The fused feature vector F fused or the feature vector of a single modality F is input into the trained disease detection model, and the probability map of each pixel belonging to the lesion area is output, expressed as:

[0059] P(x,y)=σ(W·F+b);

[0060] Where P(x,y) is the probability of disease at pixel position (x,y), sigma is the Sigmoid activation function, W is the model weight matrix, F is the input feature vector, and b is the bias term;

[0061] The non-maximum suppression (NMS) technique is applied to remove redundant bounding boxes and retain the most representative lesion area candidate boxes;

[0062] According to the lesion area predicted by the model and its corresponding probability, a preliminary diagnosis report is generated.

[0063] As a preferred scheme of the artificial intelligence-based medical image intelligent analysis system described in the present application, wherein: the preliminary diagnosis result obtained by the diagnosis module is analyzed using an attention mechanism or other explanatory method to obtain a visual report, the specific steps being:

[0064] The preliminary diagnosis result generated by the diagnosis module is input;

[0065] The preliminary diagnosis result includes the location, type and severity assessment of the lesion area;

[0066] The Grad-CAM technique is used to generate a heat map to highlight the key areas that the model focuses on when making a diagnosis decision;

[0067] Superimpose the generated heat map with the original medical image to form an intuitive visualization effect;

[0068] Add detailed explanatory notes in the visualization report;

[0069] The explanatory notes include:

[0070] The specific location and size of the lesion area;

[0071] The type and probability value of the lesion predicted by the model;

[0072] Severity assessment and recommendations based on clinical knowledge;

[0073] Integrate all the above elements into a comprehensive visualization table.

[0074] As a preferred scheme of the artificial intelligence-based medical image intelligent analysis system of the present application, the Grad-CAM technology is used to generate a heat map to highlight the key areas of interest in the model when making a diagnosis decision, and the specific steps are as follows:

[0075] For each lesion area, calculate its corresponding gradient information and backpropagate it to the last convolutional layer;

[0076] Calculate the weighted activation map, the expression is:

[0077]

[0078] where L c is the activation map of the c-th class, is the weight of the i-th feature map with respect to class c, and A i is the i-th feature map.

[0079] Apply the ReLU function to filter out negative values and ensure that the heat map only emphasizes areas that positively contribute to classification, the expression is:

[0080] R(L c )=max(0,L c );

[0081] where R is the ReLU activation function used to generate the final heat map.

[0082] In a second aspect, the present application provides a computer device comprising a memory and a processor, the memory storing a computer program, wherein the computer program is executed by the processor to implement any step of the artificial intelligence-based medical image intelligent analysis system according to the first aspect of the present application.

[0083] In a third aspect, the present application provides a computer readable storage medium having stored thereon a computer program, wherein the computer program, when executed by a processor, implements any step of the artificial intelligence-based medical image intelligent analysis system according to the first aspect of the present application.

[0084] The present application has the beneficial effects that: by adding detailed metadata information to the original medical image, the effect of enhancing data management and tracking ability is achieved, providing necessary background information for subsequent data analysis, which helps to improve the accuracy of diagnosis, and by performing gray scale transformation and data enhancement operations on the image, consistent adjustment of the image is achieved, which helps to eliminate the differences caused by different imaging devices, so that the model can learn the lesion characteristics more effectively, the weighted fusion strategy is adopted to combine the feature vectors of different modalities to generate a comprehensive feature set, which improves the accuracy of disease detection, the strategy fully utilizes the advantages of multi-modal information, enhances the recognition ability of the model for complex pathological conditions, superimposes the heat map and the original medical image to form an intuitive visual effect, and adds detailed explanatory annotations, which achieves the goal of helping doctors to quickly understand and evaluate the disease, simplifies the clinical decision-making process, and improves the diagnosis efficiency and accuracy. BRIEF DESCRIPTION OF DRAWINGS

[0085] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0086] Figure 1 The schematic diagram of the artificial intelligence-based medical image intelligent analysis system in embodiment 1. DETAILED DESCRIPTION

[0087] In order to make the above-mentioned purposes, features and advantages of the present application more apparent and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the drawings of the specification.

[0088] In the following description, many specific details are set forth in order to provide a thorough understanding of the present application, but the present application can also be implemented in other ways different from those described herein, and those skilled in the art can make similar generalizations without departing from the scope of the present application, therefore the present application is not limited to the specific embodiments disclosed below.

[0089] Second, the "one embodiment" or "an embodiment" referred to herein means a particular feature, structure, or characteristic including an implementation that can be included in at least one implementation of the application. The appearances of "in one embodiment" or "in an embodiment" in various places in the specification are not necessarily all referring to the same embodiment, nor are they necessarily all referring to a single, special implementation of the application.

[0090] Embodiment 1, Reference Figure 1 As a first embodiment of the present application, the embodiment provides an artificial intelligence-based medical image intelligent analysis system, comprising:

[0091] a data acquisition module, a data processing module, a multi-modal fusion module, a model training module, a diagnosis module, and a visualization module.

[0092] The data acquisition module is configured to acquire medical images from a plurality of medical imaging devices and obtain raw medical image data.

[0093] Further, the standard interface DICOM protocol is connected to the hospital information system or the picture archiving and communication system to extract medical images that meet the requirements.

[0094] Each image is accompanied by metadata information when collected to ensure the integrity and traceability of the data.

[0095] The metadata information includes patient ID, acquisition time, and device model.

[0096] For images that do not pass the quality assessment, a contrast stretching technique is used to improve the gray scale dynamic range of the images and obtain a raw medical image data set.

[0097] The data processing module is configured to preprocess the raw medical image data using data cleaning, normalization, and enhancement techniques to obtain a high-quality image data set.

[0098] Further, the metadata information of each image is checked, and duplicate or incomplete image data is deleted.

[0099] For images containing obvious errors or not meeting the collection standards, they are marked and removed.

[0100] The image files that may be damaged are detected and repaired using the check algorithm MD5.

[0101] All images are converted to grayscale images, and the intensity values of each pixel are located within a specific range through linear transformation to eliminate differences between different imaging devices.

[0102] The images are subjected to data enhancement.

[0103] Data enhancement includes rotation, flipping, and scaling operations.

[0104] During the training phase, an appropriate amount of artificial noise is added to the image, and a filter is used to remove the actual noise.

[0105] After the above steps, all images are saved in a uniform format, forming a high-quality image dataset.

[0106] The multimodal fusion module is used to integrate high-quality image datasets using data fusion methods, combining data from different types of imaging devices to generate a comprehensive feature set;

[0107] Furthermore, for each type of medical image, a convolutional neural network (CNN) is used to extract key features from the image;

[0108] Rigid body transformation is used to match the same anatomical location in images of different modalities. The expression is:

[0109] T(x,y) = A·x + B·y + C;

[0110] Where T is the spatial transformation function, and A, B and C are transformation matrix parameters;

[0111] After spatial registration is completed, a weighted fusion strategy is used to combine feature vectors from different modalities to generate a comprehensive feature set, expressed as:

[0112] F fused =α·F CT +(1-α)·F MRI ;

[0113] Among them, F fused F is the fused feature vector. CT and F MRI are the feature vectors of CT and MRI, respectively, and α is the weighting parameter;

[0114] The feature vectors, after feature extraction, spatial registration, and feature fusion, are stored to form a comprehensive feature set.

[0115] The model training module is used to train the comprehensive feature set using deep learning algorithms, and improve the model performance by adjusting model parameters and optimizing the loss function to obtain the disease detection model;

[0116] Furthermore, ResNet50 was chosen as the base model;

[0117] The comprehensive feature set is divided into a training set, a validation set, and a test set;

[0118] Using the cross-entropy loss function as the training objective function L, the expression is:

[0119]

[0120] where L is the loss value, y i is the true label, is the predicted value, and N is the number of samples.

[0121] The comprehensive feature vector in the training set is input into the initialized model for training, and the loss function defined above is minimized, expressed as:

[0122]

[0123] where θ is the model parameter, η is the learning rate, and t represents the number of iterations.

[0124] Apply the early stopping strategy ES to stop training when the performance on the validation set no longer improves, to avoid overfitting the training data.

[0125] After the model reaches optimal performance on the validation set, perform comprehensive evaluation on an independent test set to ensure the model has good generalization ability.

[0126] After evaluation, deploy the model to the production environment for actual disease detection work.

[0127] Diagnosis module, used to analyze new medical images using disease detection model, identify and locate possible lesion areas, and get preliminary diagnosis results.

[0128] Further, import the new medical image to be analyzed into the system through the DICOM standard interface.

[0129] Perform the same preprocessing steps as in the training phase on the newly input image, including size adjustment, grayscale standardization, and necessary data enhancement operations.

[0130] Use the same deep learning model architecture as in the training process to extract feature vectors from the preprocessed new image.

[0131] When the new image comes from multiple imaging devices, then the feature vectors of these different modalities need to be fused.

[0132] Use a weighted fusion strategy to generate a comprehensive feature vector, expressed as:

[0133] F fused = α·F CT + (1-α)·F MRI ;

[0134] where F fused is the fused feature vector, F CT and F MRIThe feature vectors of CT and MRI, respectively, and a weight parameter α is used to dynamically adjust the importance of the modalities;

[0135] The fused feature vector F fused or the feature vector F of a single modality is input into the trained disease detection model, and a probability map of each pixel belonging to a lesion area is output, expressed as:

[0136] P(x,y)=σ(W·F+b);

[0137] where P(x,y) is the probability of a lesion at pixel position (x,y), σ is the Sigmoid activation function, W is the model weight matrix, F is the input feature vector, and b is the bias term.

[0138] The non-maximum suppression (NMS) technique is applied to remove redundant bounding boxes and retain the most representative lesion area candidate boxes.

[0139] According to the lesion area predicted by the model and its corresponding probability, a preliminary diagnosis report is generated.

[0140] The visualization module is used to analyze the preliminary diagnosis results obtained by the diagnosis module using attention mechanisms or other interpretive methods to obtain a visual report.

[0141] Further, the preliminary diagnosis results generated by the diagnosis module are used as input.

[0142] The preliminary diagnosis results include the location, type, and severity assessment of the lesion area.

[0143] The Grad-CAM technique is used to generate a heat map to highlight the key areas that the model focuses on when making a diagnosis decision.

[0144] For each lesion area, its corresponding gradient information is calculated and backpropagated to the last convolutional layer.

[0145] The weighted activation map is calculated, expressed as:

[0146]

[0147] where L c is the activation map of the c-th class, is the weight of the i-th feature map with respect to class c, and A i is the i-th feature map.

[0148] The ReLU function is applied to filter out negative values, ensuring that the heat map only emphasizes areas that positively contribute to classification, expressed as:

[0149] R(L c )=max(0,Lc );

[0150] wherein R is a ReLU activation function, used to generate the final heat map;

[0151] The generated heat map is superimposed with the original medical image to form an intuitive visualization effect;

[0152] Detailed explanatory annotations are added in the visualization report;

[0153] The explanatory annotations include:

[0154] The specific location and size of the lesion area;

[0155] The type and probability value of the lesion predicted by the model;

[0156] Severity assessment and recommendations based on clinical knowledge;

[0157] All the above elements are integrated into a comprehensive visualization report.

[0158] The embodiment also provides a computer device suitable for the case of the medical image intelligent analysis system based on artificial intelligence, including a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute the computer executable instructions to realize the medical image intelligent analysis system based on artificial intelligence as proposed in the above embodiment.

[0159] The computer device can be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium to run. The communication interface of the computer device is used to communicate with external terminals in a wired or wireless manner. The wireless manner can be achieved through Wi-Fi, operator network, NFC (near field communication) or other technologies. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer overlaid on the display screen, or a key, trackball or touchpad arranged on the shell of the computer device. In addition, the input device can also be an external keyboard, touchpad or mouse, etc.

[0160] The embodiment also provides a storage medium, which stores a computer program, and the computer program is executed by a processor to implement the medical image intelligent analysis system based on artificial intelligence proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, a magnetic disk or an optical disk.

[0161] To sum up, by adding detailed metadata information to the original medical image, the effect of enhancing data management and tracking ability is achieved, necessary background information is provided for subsequent data analysis, which helps to improve the accuracy of diagnosis, the consistency of the image is adjusted by the gray scale transformation and data enhancement operation, which helps to eliminate the difference caused by different imaging devices, so that the model can learn the lesion characteristics more effectively, the weighted fusion strategy is adopted to combine the feature vectors of different modalities to generate a comprehensive feature set, which improves the accuracy of disease detection, the strategy fully utilizes the advantages of multi-modal information, enhances the recognition ability of the model for complex pathological conditions, the heat map is superimposed on the original medical image to form an intuitive visual effect, and detailed explanatory annotations are added, which achieves the goal of helping doctors to quickly understand and evaluate the disease, and simplifies the clinical decision-making process, while improving the diagnosis efficiency and accuracy.

[0162] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present application but not limit the present application, although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced equivalently without departing from the spirit and scope of the technical solutions of the present application, and all of them should be covered in the scope of the claims of the present application.

Claims

1. An artificial intelligence-based medical image intelligent analysis system, characterized by: The application relates to a medical image diagnosis system based on multi-modal fusion and deep learning. The system comprises a data acquisition module, a data processing module, a multi-modal fusion module, a model training module, a diagnosis module, and a visualization module. The data acquisition module is used for acquiring medical images from multiple medical imaging devices and obtaining original medical image data. The data processing module is used for pre-processing the original medical image data by using data cleaning, normalization, and enhancement techniques to obtain a high-quality image dataset. The multi-modal fusion module is used for integrating the high-quality image dataset by using a data fusion method to combine data from different types of imaging devices to generate a comprehensive feature set. The model training module is used for training the comprehensive feature set by using a deep learning algorithm, adjusting model parameters, and optimizing loss functions to improve model performance to obtain a disease detection model. The diagnosis module is used for analyzing new medical images by using the disease detection model to identify and locate possible lesion areas to obtain preliminary diagnosis results. The visualization module is used for analyzing the preliminary diagnosis results obtained by the diagnosis module by using an attention mechanism or other explanatory methods to obtain a visual report. 2.The artificial intelligence-based medical image intelligent analysis system of claim 1, wherein: The specific steps of acquiring medical images from multiple medical imaging devices and obtaining original medical image data are as follows: Connect the hospital information system or the picture archiving and communication system through the standardized interface DICOM protocol to extract medical images meeting the requirements; Each image is attached with metadata information to ensure the integrity and traceability of the data; The metadata information includes patient ID, acquisition time, and device model; For images that do not pass the quality assessment, the contrast stretching technique is used to improve the gray scale dynamic range of the images to obtain the original medical image dataset. 3.The artificial intelligence-based medical image intelligent analysis system of claim 2, wherein: The specific steps of pre-processing the original medical image data by using data cleaning, normalization, and enhancement techniques to obtain a high-quality image dataset are as follows: Check the metadata information of each image, and delete repeated or incomplete image data; Mark and remove images containing obvious errors or not meeting the acquisition standards; Detect and repair possibly damaged image files by using the check algorithm MD5; Convert all images to grayscale images, and use the linear transformation method to make the intensity value of each pixel within a specific range to eliminate the differences between different imaging devices; Perform data enhancement on the images; The data enhancement includes rotation, flipping, and scaling operations; Add appropriate artificial noise to the images during the training stage, and use a filter to remove the actual existing noise; After the above steps, all images are saved in a unified format to form a high-quality image dataset. 4.The artificial intelligence-based medical image intelligent analysis system of claim 3, wherein: The specific steps of integrating the high-quality image dataset by using a data fusion method to combine data from different types of imaging devices to generate a comprehensive feature set are as follows: For each type of medical image, use a convolutional neural network (CNN) to extract key features from the image; Use rigid transformation to match the same anatomical position in different modal images, and the expression is as follows: T(x,y) = A*x + B*y + C; Wherein, T is a spatial transformation function, A, B, and C are transformation matrix parameters. After completing spatial registration, the feature vectors from different modalities are combined using a weighted fusion strategy to generate a comprehensive feature set, expressed as: F fused = a · F CT + (1 - a) · F MRI ; Where F fused is the fused feature vector, F CT and F MRI are the feature vectors of CT and MRI, respectively, and a is a weight parameter. Where F fused is the fused feature vector, F CT and F MRI are the feature vectors of CT and MRI, respectively, and a is a weight parameter. The feature vectors after feature extraction, spatial registration, and feature fusion are stored to form a comprehensive feature set. 5.The artificial intelligence-based medical image intelligent analysis system of claim 4, wherein: The comprehensive feature set is trained using a deep learning algorithm, and the model performance is improved by adjusting the model parameters and optimizing the loss function to obtain a disease detection model, with the following specific steps: ResNet50 is selected as the base model; The comprehensive feature set is divided into training set, validation set, and test set; The cross-entropy loss function is used as the target function L for training, expressed as: wherein L is a loss value, y i is a true label, is a predicted value, and N is a sample number; The comprehensive feature vectors in the training set are input into the initialized model for training, and the loss function defined above is minimized, expressed as: where θ is the model parameter, η is the learning rate, and t represents the number of iterations; The early stopping strategy ES is applied to stop training when the performance on the validation set no longer improves, avoiding overfitting of the training data; After the model reaches optimal performance on the validation set, it is comprehensively evaluated on an independent test set to ensure good generalization ability of the model; After evaluation, the model is deployed to the production environment for actual disease detection work. 6.The artificial intelligence-based medical image intelligent analysis system of claim 5, wherein: The new medical image is analyzed using the disease detection model to identify and locate possible lesion areas, obtaining preliminary diagnostic results, with the following specific steps: The new medical image to be analyzed is imported into the system through the DICOM standard interface; The same preprocessing steps as in the training phase are performed on the newly input image, including size adjustment, grayscale normalization, and necessary data enhancement operations; The same deep learning model architecture as in the training process is used to extract feature vectors from the preprocessed new image; When the new image comes from multiple imaging devices, the feature vectors from different modalities need to be fused; A weighted fusion strategy is used to generate a comprehensive feature vector, expressed as: F fused = a · F CT + (1 - a) · F MRI ; where F fused is the fused feature vector, F CT and F MRI are the feature vectors of CT and MRI respectively, and a is a weight parameter that dynamically adjusts according to the importance of the modalities. The fused feature vector F fused or the feature vector F of a single modality is input into the trained disease detection model, and a probability map of each pixel belonging to the lesion area is output, expressed as: P(x,y)=σ(W·F+b); where P(x,y) is the lesion probability at pixel position (x,y), σ is the Sigmoid activation function, W is the model weight matrix, F is the input feature vector, and b is the bias term; The non-maximum suppression NMS technique is applied to remove redundant bounding boxes and retain the most representative lesion area candidate boxes; According to the lesion area predicted by the model and its corresponding probability, a preliminary diagnosis report is generated. 7.The artificial intelligence-based medical image intelligent analysis system of claim 6, wherein: The preliminary diagnostic results obtained by the diagnosis module are analyzed using attention mechanisms or other explanatory methods to obtain a visual report, with the following specific steps: The preliminary diagnostic results generated by the diagnosis module are used as input; The preliminary diagnostic results include the location, type, and severity assessment of the lesion area; The Grad-CAM technique is used to generate a heat map, highlighting the key areas that the model focuses on when making a diagnosis decision; The generated heat map is superimposed on the original medical image to form an intuitive visual effect; Detailed explanatory annotations are added to the visual report; The explanatory annotations include: The specific location and size of the lesion area; The lesion type and probability value predicted by the model; Severity assessment and recommendations based on clinical knowledge; Integrate all the above elements into a comprehensive visual report. 8.The artificial intelligence-based medical image intelligent analysis system of claim 7, wherein: The heat map is generated using the Grad-CAM technique, highlighting the key areas that the model focuses on when making diagnostic decisions. The specific steps are as follows: For each lesion area, calculate its corresponding gradient information and backpropagate it to the last convolutional layer. Calculate the weighted activation map, the expression is: where L c is the activation map of the c-th class, is the weight of the i-th feature map with respect to class c, A i is the i-th feature map; Apply the ReLU function to filter out negative values, ensuring that the heat map only emphasizes areas that positively contribute to classification, the expression is: R(L c ) = max(0, L c ) ; Where R is the ReLU activation function, used to generate the final heat map.

Images