Coronary artery CTA epicardial fat automatic extraction and quantification method based on artificial intelligence

Abstract

The invention discloses a coronary artery CTA epicardial fat automatic extraction and quantification method based on artificial intelligence, and aims to improve the accuracy and efficiency of fat extraction. The method comprises the following steps: preprocessing a coronary artery CTA image to enhance data quality; a segmentation model based on U-Net is constructed, and accurate segmentation of the heart structure is achieved; the epicardial adipose tissues in the segmented regions are automatically extracted in combination with a set CT value threshold value; quantifying the fat volume by adopting a voxel counting mode, and analyzing the density of the fat volume; a three-dimensional visual model is further generated, and the spatial relation between fat distribution and the heart structure is displayed. According to the method, the effect is verified on multiple data sets, it is shown that the method is superior to an existing method in the aspects of fat extraction precision, volume and density evaluation and model robustness, and the method has good application prospects.

Description

Technical Field

[0001] This invention relates to the field of intelligent healthcare, specifically to a method for automatically extracting and quantitatively analyzing epicardial fat from coronary CTA (computed tomography angiography) images using deep learning technology. This technology, by combining image preprocessing, automatic cardiac structure recognition and segmentation, epicardial fat detection, and precise fat mass calculation, can effectively assist physicians in the diagnosis and assessment of cardiovascular disease, providing more accurate and rapid analysis results. This technology utilizes an improved convolutional neural network model, U-Net, to process and analyze medical images, integrating modern computing technologies to enhance the processing capabilities and application efficiency of medical imaging. Background Art

[0002] In modern medical diagnosis, coronary computed tomography angiography (CTA) is an important tool for diagnosing and evaluating coronary artery disease. Epicardial fat, the fatty tissue located between the heart's surface and the pericardium, has recently been found to be significantly associated with the risk of various cardiovascular diseases. Therefore, accurately measuring the distribution and amount of epicardial fat is an important factor in assessing cardiovascular risk.

[0003] Traditional methods for assessing epicardial fat rely primarily on manual measurement by physicians through visual inspection of CTA images. This method is not only time-consuming and inefficient, but the results are also significantly influenced by the operator's skill and experience, and are highly subjective. With technological advances, computer-assisted image analysis methods have begun to be applied to the automated processing of medical images. However, early methods were often limited by image resolution, contrast, and algorithmic complexity, making it difficult to accurately segment cardiac structures and identify epicardial fat.

[0004] In recent years, deep learning techniques, particularly convolutional neural networks (CNNs), have demonstrated strong potential in medical image analysis. Network architectures such as U-Net have been widely used for automated medical image segmentation due to their excellent image segmentation performance. These networks can automatically identify and segment the heart and its associated structures by learning from large amounts of annotated data, thereby enabling the automated extraction and quantification of epicardial fat.

[0005] However, applying these advanced image analysis techniques to the automatic extraction and quantitative analysis of epicardial fat still faces challenges. For example, the complex anatomical structure of the heart, noise and artifacts in CTA images may affect the accurate identification and segmentation of adipose tissue. In addition, existing deep learning models usually require a large amount of labeled data for training, and the model's generalization ability and adaptability to new data are also hot issues in current research. Therefore, developing a method that integrates the latest artificial intelligence technologies to achieve efficient and accurate automatic extraction and quantitative analysis of epicardial fat is an urgent problem that needs to be solved in this technical field.

[0006] Technical Solution

[0007] This paper proposes an AI-based method for the automated extraction and quantification of epicardial fat from coronary CTA. This technical solution aims to improve the automation and accuracy of epicardial fat extraction and quantification by integrating advanced image preprocessing, deep learning models, automated detection and segmentation algorithms, and efficient post-processing techniques. The technical solution specifically includes the following key steps:

[0008] Preferably, in the multimodal image preprocessing stage, the present invention adopts several advanced technologies to further improve the accuracy and efficiency of epicardial fat analysis in coronary CTA. First, nonlinear noise filtering uses an algorithm based on non-local means or Gaussian mixture models, which is more effective than traditional filtering techniques in suppressing random noise while maintaining the clarity of image details. Next, contrast enhancement is performed using adaptive histogram equalization or contrast-limited adaptive histogram equalization technology to enhance the contrast between cardiac structures and surrounding tissues, thereby improving image recognition. In addition, normalization processing standardizes all image data to a uniform numerical range to ensure that the deep learning model receives consistent input data. Finally, image de-artifacting processing uses a deep learning-based method to eliminate artifacts caused by scanning equipment or patient movement, thereby ensuring the accuracy of subsequent processing steps.

[0009] Preferably, in the deep learning-based cardiac structure segmentation stage, the present invention utilizes an improved U-Net architecture, which is specifically optimized for the complexity and variability of cardiac structure. The improved U-Net includes deeper network layers and enhanced feature extraction capabilities, using depthwise separable convolutional layers to reduce model parameters and computational burden while maintaining efficient information processing capabilities.

[0010] Preferably, spatial feature extraction: U-Net performance is further enhanced by integrating positional encoding. This positional encoding provides additional information about the location of pixels in the image, helping the model better understand the spatial structure and patterns of the heart and surrounding tissues.

[0011] Preferably, a feature fusion strategy is designed to effectively integrate spatial features extracted by the U-Net model with contextual features extracted from other models (such as a pre-trained CNN). This strategy includes operations such as feature mapping, weighting, and pooling to enhance the expressiveness and robustness of features.

[0012] Preferably, deep network architecture design: a deep network architecture is constructed, which contains multiple improved U-Net layers, each layer can process and refine the features from the previous layer, gradually improving the abstraction level of the features to adapt to the complex cardiac structure segmentation task.

[0013] Preferably, define a loss function: Define a loss function that measures the difference between the model's predictions and the true labels. This loss function is specifically designed to emphasize the accuracy of the boundaries of the heart and epicardial fat regions. Its choice directly affects the model's training and segmentation performance.

[0014] Optimization algorithm selection: Choose an appropriate optimization algorithm, such as Adam or SGD, to minimize the loss function and update the model parameters. This may also include a learning rate adjustment strategy to ensure stability and efficiency during model training.

[0015] Optimally, model training and validation are performed using a large, well-annotated coronary CTA image dataset. During training, techniques such as cross-validation are used to assess the model's generalization capabilities, ensuring it is not overfitting and can provide reliable segmentation results in a variety of clinical scenarios.

[0016] Preferably, segmentation decision making: After model training is complete, the trained deep U-Net model is used to identify and segment epicardial fat in new coronary CTA image data. Segmentation decisions are based on the pixel-level segmentation probability map output by the model, which includes thresholding to determine which regions have a high enough probability to be identified as epicardial fat. Furthermore, this involves merging and partitioning consecutive fat regions into separate segmented entities.

[0017] Post-processing and optimization: The segmentation results are post-processed to further improve segmentation accuracy and practicality. This post-processing step includes applying image denoising techniques, smoothing to remove segmentation noise, and edge preservation techniques to maintain a clear boundary between fat and non-fat tissue. Furthermore, optimization techniques such as non-maximum suppression (NMS) are used to refine and precisely define the boundaries of the fat region, ensuring its continuity and integrity.

[0018] Optimally, model deployment and application: This invention involves deploying a trained deep learning model into a real-world medical image processing system to achieve automated epicardial fat extraction and quantification. This model is specifically designed to accommodate diverse clinical applications, such as cardiovascular disease risk assessment, pre-operative surgical planning, and daily health monitoring.

[0019] Preferably, in the fat volume calculation stage: the total fat volume is obtained by calculating the number of voxels in the segmented fat region and multiplying the number by the volume of a single voxel. The calculation formula is:

[0020]

[0021] in, is the total volume of epicardial fat, is the number of voxels labeled as fat, is the volume of a single voxel. The calculation formula is:

[0022]

[0023] in, Refers to the actual physical width and height (in millimeters) of each pixel in the CT scan, obtained in the DICOM file, Refers to the thickness of each slice (in millimeters), which can also be obtained from the DICOM file.

[0024] Preferably, during the fat density analysis phase, statistical analysis of pixel intensity is used to depict the uniformity and aggregation characteristics of fat tissue. The calculation formula is:

[0025]

[0026]

[0027] in, It is The pixel intensity of each voxel labeled as fat, is the average pixel intensity of the fat region, is the standard deviation of pixel intensity in the fat region, which is used to measure the uniformity of density.

[0028] Preferably, a three-dimensional reconstruction technique is used to convert the two-dimensional segmented image into a three-dimensional model, which not only helps doctors understand the spatial distribution of epicardial fat more intuitively, but also allows comparison with other cardiac structures to evaluate its potential clinical impact.

[0029] The technical solution of the present invention realizes a novel artificial intelligence-based method for automatic extraction and quantification of epicardial fat in coronary CTA through the above steps. This method can not only automatically identify and segment epicardial fat, but also effectively perform quantitative analysis, significantly improving the accuracy and efficiency of processing.

[0030] This invention provides an innovative solution to the technical challenges in medical image processing. First, through an improved U-Net model and deep learning techniques, it effectively captures the complex features of cardiac and adipose tissue in images, overcoming the limitations of traditional image processing methods in accurately segmenting epicardial fat. Second, to address the high resolution and rich details of cardiac CTA images, this invention employs efficient multimodal image preprocessing techniques to reduce noise while retaining key information, enhancing the model's ability to process detail. Furthermore, this invention designs an innovative feature fusion strategy that effectively integrates spatial and contextual features, improving the expressiveness and robustness of features. Regarding model training and optimization, a carefully designed network architecture and loss function, combined with advanced optimization algorithms, enhances the efficiency and stability of model training. Furthermore, to improve the model's generalization, this invention employs techniques such as cross-validation to ensure the model's adaptability to new data. Finally, through post-processing techniques and quantitative analysis optimization strategies, the accuracy and consistency of quantitative results are further improved, meeting the needs of clinical diagnosis and treatment planning. This provides an efficient and accurate method for automatic epicardial fat extraction and quantification in the field of medical image analysis.

[0031] The innovation of the present invention lies in two core protection points: First, the present invention uses advanced deep learning models, especially the improved U-Net architecture, and is specially designed for the automatic identification and precise segmentation of epicardial fat. Compared with traditional manual or semi-automatic image segmentation methods, this automated processing significantly improves operational efficiency and the repeatability of results. Secondly, the present invention designs a voxel counting method to perform volume quantification and density analysis on the identified epicardial fat, providing doctors with quantitative data support, which is often difficult to accomplish automatically with traditional methods. These beneficial effects work together to make the present invention have important application value and broad market prospects in the field of medical image analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0033] Figure 1 This is a flow chart of an artificial intelligence-based method for the automatic extraction and quantification of epicardial fat in coronary CTA;

[0034] Figure 2 A schematic diagram of the epicardial fat segmentation process according to an embodiment of the present invention;

[0035] Figure 3 The heart segmentation model training results and evaluation indicators provided by the embodiment of the present invention;

[0036] Figure 4 A schematic diagram of a heart segmentation result provided by an embodiment of the present invention;

[0037] Figure 5 This is a schematic diagram of the epicardial fat segmentation results provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0038] The specific embodiment of the present invention describes in detail the experimental process, parameter configuration and experimental results analysis of the artificial intelligence-based coronary CTA epicardial fat automatic extraction and quantification method to demonstrate the effectiveness and superiority of the proposed method.

[0039] S1. Experimental environment configuration

[0040] The experiments were conducted on a server equipped with an NVIDIA GeForce RTX 3070 GPU, an Intel Core i9 processor, 64GB of RAM, and Windows 11. PyTorch 2.0.0 was used as the deep learning framework, and CUDA 11.2 and cuDNN 8.2 were installed to optimize GPU acceleration performance.

[0041] S2. Detailed steps of data preprocessing

[0042] The coronary CTA image dataset used in the experiment consists of 218 cardiac slice images annotated by professional physicians. Data preprocessing includes two main steps: normalization and noise removal:

[0043] a. Normalization: To eliminate dimensionality differences between bands and standardize the data input format, the pixel values in each band undergo a linear transformation and are scaled to the range [0, 1]. This is achieved by subtracting the band's minimum value from the raw intensity value of each pixel and then dividing it by the difference between the band's maximum and minimum values. Normalization is a key preprocessing step that helps accelerate model convergence and improve generalization.

[0044] b. Noise Removal: To reduce random noise and improve image quality, a 3x3 Gaussian filter was applied to each band. The standard deviation σ of the Gaussian kernel was set to 0.5, which helps smooth the image without excessively blurring important anatomical structures. This Gaussian filter not only reduces Gaussian noise but also preserves image detail.

[0045] After preprocessing, the image dataset was randomly divided into a training set (60%), a validation set (20%), and a test set (20%).

[0046] S3, deep U-net model architecture parameters

[0047] This paper employs an improved deep U-Net model architecture, specifically optimized for the automatic extraction and quantitative analysis of epicardial fat from coronary CTA. This model consists of a symmetrical encoder and decoder. The encoder downsamples through successive convolutional layers (each followed by ReLU activation and batch normalization) and max pooling steps, doubling the number of feature maps starting at 64. The decoder upsamples through transposed convolutions, gradually halving the number of feature maps. Convolution is performed after each upsampling step to refine and restore image detail. Furthermore, skip connections, a key feature of U-Net, pass high-resolution feature maps from the encoder directly to the corresponding layers of the decoder, helping to restore detail and improve segmentation accuracy. Finally, 1x1 convolutions are used to output accurate segmentation results. The application of this deep learning framework not only significantly improves the accuracy of epicardial fat analysis but also optimizes processing efficiency.

[0048] S4. Feature extraction and fusion strategy

[0049] In the present invention, an advanced feature extraction and fusion strategy is adopted to improve the accuracy and robustness of the automatic extraction and quantitative analysis of epicardial fat in coronary CTA. Through the improved U-Net model, this strategy realizes multi-scale feature extraction, in which the encoder captures a wide range of contextual information through layer-by-layer downsampling, while the decoder restores detail information and spatial accuracy through upsampling and skip connections, and comprehensively extracts image features from fine to coarse granularity. In addition, feature fusion directly combines the high-resolution feature maps in the encoder with the corresponding layers of the decoder through skip connections, which not only helps to restore the lost detail information, but also fuses features at different levels to enhance the model's recognition ability of boundaries and textures. Advanced fusion blocks and attention mechanisms are further adopted in the decoder to optimize the final combination of feature maps from different layers, ensuring that the output segmentation map is optimal in terms of accuracy and coherence. These strategies work together to significantly improve processing efficiency and the reliability of results.

[0050] S5. Classifier design and parameters

[0051] The classifier is built on two fully connected layers, used to parse features extracted from a deep U-Net model and perform final recognition and quantification. The number of nodes in the first fully connected layer is 2048, and the number of nodes in the second fully connected layer matches the number of categories in the classification task. ReLU is used as the activation function for the intermediate layers, and the output layer uses either a Sigmoid or Softmax function depending on the task requirements. Since this invention is used for binary classification, the Sigmoid function is selected.

[0052] S6. Model training and optimization algorithm

[0053] The model was trained using the cross-entropy loss function and the Adam optimization algorithm with parameters β1=0.9, β2=0.999, and ε=10^-8. The initial learning rate was set to 0.001 and was decayed by 0.1 every five epochs. Early stopping was used to control overfitting; training was terminated if the validation set loss did not decrease after 10 consecutive epochs.

[0054] S7, post-processing technology application

[0055] The classification results are post-processed using non-maximum suppression (NMS) to improve the accuracy and robustness of the classification results. The NMS confidence threshold is set to 0.3 and the neighborhood size is 3x3 pixels to remove pixels with low classification confidence and preserve the classification boundaries.

[0056] S8. Quantitative calculation

[0057] a. Volume quantification: The total fat volume is obtained by counting the number of voxels within the segmented fat region and multiplying it by the volume of a single voxel. The formula is:

[0058]

[0059] in, is the total volume of epicardial fat, is the number of voxels labeled as fat, is the volume of a single voxel, calculated as:

[0060]

[0061] in, Refers to the actual physical width and height (in millimeters) of each pixel in the CT scan, obtained in the DICOM file, Refers to the thickness of each slice (in millimeters), which can also be obtained from the DICOM file.

[0062] b. Density analysis: Evaluate the pixel density distribution within the fat area and use statistical analysis of pixel intensity to depict the uniformity and aggregation characteristics of fat tissue. The formula is:

[0063]

[0064]

[0065] in, It is The pixel intensity of each voxel labeled as fat, is the average pixel intensity of the fat region, is the standard deviation of pixel intensity in the fat region, which is used to measure the uniformity of density.

[0066] S9. Evaluation and analysis of experimental results

[0067] The present invention uses recall rate (Recall), mean pixel accuracy rate (Mean Pixel Accuracy, MPa), precision rate (Precision), mean intersection over Union (mIoU) as evaluation indicators for experiments. After 200 epochs of training. The model designed by the present invention performs excellently in key performance indicators such as recall rate, mean pixel accuracy rate, precision rate and mean intersection over Union rate on the test set, reaching 99.15%, 99.15%, 99.12% and 98.29% respectively, showing the significant technical advantages of the model, and also indicating its application potential for cardiac segmentation in coronary CTA images, with significant commercial value and broad application prospects. The training results of the cardiac segmentation model are as follows: Figure 3 As shown in the figure, the heart segmentation results are as follows Figure 4 As shown in the figure, the epicardial fat segmentation results are as follows Figure 5 shown.

[0068] S10. Conclusion

[0069] The present invention effectively achieves the automatic extraction and quantitative analysis of epicardial fat in coronary CTA by adopting an improved deep U-Net model structure. This method combines multi-scale feature extraction, advanced feature fusion strategies, and advanced activation functions to significantly improve the accuracy and efficiency of epicardial fat analysis. Through these technologies, the present invention can not only accurately identify and quantify epicardial fat, but also optimize the processing flow, improve the generalization ability and practicality of the model. In addition, the implementation of the present invention provides a powerful tool for supporting the diagnosis and treatment planning of cardiovascular diseases, and demonstrates broad application prospects in the field of medical image analysis. Ultimately, this efficient and reliable automated analysis method has significant potential in improving the efficiency and accuracy of clinical decision-making, while providing rich data support and analytical depth for related research. These results show that the present invention is not only technologically innovative, but also demonstrates great value in practical applications, helping to promote the advancement of medical image processing technology and the development of the medical and health field.

Claims

1. A method for automatic extraction and quantification of epicardial fat in coronary CTA based on artificial intelligence, characterized in that: The method comprises the following steps: a. Multimodal Image Preprocessing: First, the acquired coronary CTA image data undergoes advanced preprocessing, including nonlinear noise filtering, contrast enhancement, and normalization, to optimize image quality and prepare the data in a format suitable for deep learning model processing. b. Cardiac Structure Segmentation Based on Deep Convolutional Neural Networks: A modified U-Net architecture is applied, which enhances the model's ability to identify cardiac tissue boundaries by incorporating depthwise separable convolutions and batch normalization. This step aims to accurately segment cardiac structures from CTA images, providing a foundation for localizing epicardial fat regions. c. Automatic Detection and Segmentation of Epicardial Fat Regions: Threshold analysis techniques combined with morphological operations are used to perform detailed fat detection on the cardiac regions segmented in the previous step. Adaptive threshold adjustment and region growing algorithms are used to accurately identify and isolate epicardial fat tissue. d. Quantitative Analysis and Fat Mass Calculation: A voxel counting method is designed to quantify the volume of identified epicardial fat. Statistical analysis techniques are also used to assess the uniformity and density of fat distribution, providing a quantitative basis for subsequent pathological analysis and risk assessment. e. High-dimensional data visualization and output: The processing results are displayed using high-dimensional data visualization technology, including a 3D reconstructed heart model and fat distribution map. Detailed analysis reports and quantitative data are also output to support clinical decision-making and medical research.

2. The method according to claim 1, characterized in that The multimodal image preprocessing steps include nonlinear noise filtering, contrast enhancement, normalization and image artifact removal. Nonlinear noise filtering uses algorithms based on non-local mean or Gaussian mixture models to effectively suppress random noise in CTA images while maintaining detail clarity. Contrast enhancement improves the contrast between cardiac structures and surrounding tissues through adaptive histogram equalization or contrast-limited adaptive histogram equalization technology, improving image recognition. Normalization processing standardizes image data to a uniform numerical range to ensure that the input of subsequent deep learning models has a consistent data distribution. Image artifact removal uses a deep learning-based method to eliminate artifacts caused by scanning equipment or patient movement, ensuring the accuracy of subsequent processing steps. These processing steps together improve the accuracy and robustness of cardiac structure segmentation and epicardial fat extraction, providing a solid foundation for high-quality medical image analysis.

3. The method according to claim 1, characterized in that The cardiac structure segmentation step, based on a deep convolutional neural network, is performed using an improved U-Net model. This model structure is optimized to enhance the recognition and accuracy of the heart and its associated structures. The improved U-Net includes depthwise separable convolutional layers, which improve the model's computational efficiency by reducing the number of parameters while maintaining or improving performance. Furthermore, batch normalization and activation layers are integrated into the model to stabilize the training process and accelerate convergence. The model training process involves supervised learning using a large amount of well-labeled coronary CTA image data to ensure that the model can accurately identify and segment cardiac regions. This step is critical for achieving automatic epicardial fat extraction and quantitative analysis, as accurate cardiac structure segmentation directly affects the effectiveness and accuracy of subsequent fat identification and quantification.

4. The method according to claim 1, wherein The automated epicardial fat region detection and segmentation step utilizes advanced image processing algorithms and machine learning techniques to perform a detailed analysis of the previously segmented cardiac structures. This step first applies an adaptive thresholding technique based on pixel intensity to quantitatively identify pixels within the cardiac region that are characteristic of adipose tissue. The threshold is set based on the typical Hounsfield unit (HU) range of adipose tissue, ensuring highly accurate separation of adipose and non-adipose tissue. Furthermore, the method employs morphological filtering techniques, such as opening and closing operations, to optimize the geometric continuity and boundary definition of adipose tissue. These operations help eliminate noise and false-positive regions introduced by thresholding, enhancing the identification of true adipose tissue. Advanced image processing steps also utilize statistical analysis and machine learning classification algorithms, such as support vector machines (SVMs) or deep learning convolutional neural networks, to finely distinguish adipose tissue from adjacent structures, enhancing the accuracy and reliability of the segmentation results. This multi-layered, multi-technique approach ensures that the automated epicardial fat region detection and segmentation step maintains high accuracy and low false-positive rates, while providing stable and reliable input data for subsequent quantitative analysis and clinical applications.

5. The method according to claim 1, wherein The quantitative analysis and fat mass calculation steps involve the use of sophisticated image quantification techniques to accurately assess the volume and distribution of epicardial fat. This process begins with the precise segmentation of the epicardial fat region obtained in the previous steps. The specific details and formulas are as follows: a. Volume quantification: The total fat volume is obtained by counting the number of voxels within the segmented fat region and multiplying it by the volume of a single voxel. The formula is: in, is the total volume of epicardial fat, is the number of voxels labeled as fat, is the volume of a single voxel, calculated as: in, Refers to the actual physical width and height (in millimeters) of each pixel in the CT scan, obtained in the DICOM file, Refers to the thickness of each slice (in millimeters) and can also be obtained from the DICOM file. b. Density analysis: Evaluates the pixel density distribution within the fat area and uses statistical analysis of pixel intensity to describe the uniformity and aggregation characteristics of fat tissue. The formula is: in, It is The pixel intensity of each voxel labeled as fat, is the average pixel intensity of the fat region, is the standard deviation of pixel intensities in the fat region, used to measure density uniformity. Through the above steps, the quantitative analysis and fat mass calculation steps of the present invention can characterize the uniformity and aggregation characteristics of adipose tissue. These analytical methods not only ensure the accuracy and reliability of measurement results, but also provide valuable biomarker information for clinical diagnosis and disease management, helping medical professionals better understand the distribution characteristics of epicardial fat.

6. The method according to claim 1, characterized in that The high-dimensional data visualization and output step involves converting the results of quantitative analysis and fat mass calculation into intuitive graphics and reports using advanced visualization techniques, allowing medical professionals to more clearly assess and understand the distribution of epicardial fat and its potential clinical significance. This process includes: a. 3D reconstruction: Using the data obtained from the epicardial fat segmentation and quantification steps, a 3D model of the heart and surrounding adipose tissue is constructed. This 3D visualization not only provides a spatial representation of fat distribution but also reveals the relative positional relationships between fat and key cardiac structures, which is particularly important for surgical planning and treatment decision-making. b. Dynamic view generation: This enables dynamic, interactive views, allowing physicians to observe cardiac structures and adipose tissue from multiple angles and at varying magnifications, providing a more comprehensive diagnostic perspective. c. Quantitative data report output: This automatically generates detailed statistical reports including key metrics such as total fat mass, volume distribution, and density measurements. These reports are presented in an easy-to-understand format, combining charts and numerical data, allowing medical professionals to quickly grasp the core insights of the analysis results. d. Data sharing and integration: This ensures that the visualization output is compatible and integrated with existing medical health record systems, enabling data sharing and remote access, thereby improving the efficiency and quality of medical services. Through these high-dimensional data visualization and output technologies, the present invention can effectively transform complex data into intuitive and easy-to-understand information, significantly improving the level of information support for clinical decision-making and the quality of overall medical services.

7. The method for automatic extraction and quantification of epicardial fat in coronary CTA based on artificial intelligence according to any one of claims 1 to 6, characterized in that: This method combines advanced multimodal image preprocessing techniques, an optimized deep convolutional neural network model, sophisticated fat detection and segmentation techniques, high-precision volume and density quantification analysis, and high-dimensional data visualization and output. It optimizes coronary CTA image quality through nonlinear noise filtering, contrast enhancement, normalization, and artifact removal. It employs an improved U-Net model with depthwise separable convolution and batch normalization to improve the accuracy and efficiency of cardiac structure segmentation. It uses threshold analysis and morphological operations to accurately identify and isolate epicardial fat. It uses voxel counting and statistical analysis techniques to accurately quantify fat volume. Finally, it applies 3D reconstruction and dynamic view generation techniques for high-dimensional data visualization, providing intuitive and detailed graphics and data reports for clinical decision-making, significantly improving the efficiency and accuracy of medical diagnosis and treatment.

8. A computer-readable storage medium storing a computer program specifically configured to perform artificial intelligence-based automated epicardial fat extraction and quantification in coronary CTA. This program implements a series of steps, including multimodal image preprocessing, cardiac structure segmentation, epicardial fat detection and segmentation, fat mass quantification analysis, and high-dimensional data visualization and output, as described in claims 1 to 6. The storage medium may be a hard drive, solid-state drive, optical disk, USB flash drive, SD card, or other non-volatile storage device. The program may be written in a high-level programming language and rely on a deep learning framework such as TensorFlow, PyTorch, Keras, or Scikit-learn. It is designed to run cross-platform and is compatible with Windows, Linux, and macOS operating systems. Furthermore, the program provides a user-friendly interface that allows users to input data, configure parameters, perform classification, and display results. It also includes detailed documentation, sample code, and error handling mechanisms. The program is also optimized for hardware accelerators such as GPUs or TPUs to improve execution efficiency. Through these characteristics, the storage medium enables the automatic extraction and quantification of epicardial fat in coronary CTA to be implemented on a variety of general-purpose computing devices, providing users with an easy-to-use and efficient medical image processing tool.

9. An electronic device equipped with a processor and memory, wherein the memory stores a specific computer program. This computer program is specifically designed to execute the artificial intelligence-based method for automatic epicardial fat extraction and quantification in coronary CTA according to any one of claims 1 to 6. When the processor runs this program stored in the memory, it can automatically process coronary CTA image data, including key steps such as multimodal image preprocessing, cardiac structure segmentation, epicardial fat detection and segmentation, fat mass quantification analysis, and high-dimensional data visualization and output. This electronic device, which can be a tablet computer, laptop computer, or a dedicated computing device, provides a portable and efficient medical image processing solution through built-in hardware and software resources. Users can easily carry this electronic device and quickly deploy epicardial fat extraction and quantification tasks when needed, providing powerful computing power and flexible application scenarios, whether in medical institutions or laboratory research. Furthermore, the device is designed with ease of use and accessibility in mind, enabling non-expert users to utilize deep learning technology for epicardial fat analysis, greatly expanding its scope of application and potential user base.

Images