Method, device and equipment for calculating cardiac coronary metabolic activity and storage medium

Abstract

The application provides a kind of heart coronary artery metabolic activity calculation method, device, equipment and storage medium, the method comprises: the target CT image data is input into the trained coronary artery key layer screening model, and the key slice containing blood vessel characteristics is obtained;Using digital image processing technology, through morphological operation, logical subtraction and PET / CT fusion analysis, the accurate positioning of right atrium and three blood vessels is realized;Using right atrial automatic positioning algorithm and SUV quantitative correction model, the target background difference is used to calculate the right atrial background SUV value and the maximum SUV value in the three blood vessel regions, and further output the corrected plaque metabolic index.The application realizes the atrium and coronary artery three blood vessels related metabolic parameters based on PET and CT image data, calculation and visualization display, solves the problems of relying on manual labeling, complicated process, low efficiency and strong subjectivity in traditional image analysis.

Description

Technical Field

[0001] This invention relates to the fields of multimodal information fusion, image processing, and computer-aided diagnostic technology based on artificial intelligence, and particularly to a method, apparatus, device, and storage medium for calculating cardiac coronary metabolic activity. Background Technology

[0002] Cardiovascular disease, particularly coronary artery disease (CAD), has become one of the leading causes of death worldwide. In clinical diagnosis and risk assessment, identifying vulnerable plaques at high rupture risk is crucial for preventing acute coronary syndromes. The combined imaging technique of positron emission tomography (PET) and computed tomography (CT), which integrates the high-resolution anatomical information of CT images with the high-sensitivity molecular metabolic information of PET images, has become a key imaging method for assessing vascular inflammatory activity, microcalcification, and plaque stability.

[0003] In PET / CT imaging analysis, the Standardized Uptake Value (SUV) is a core indicator for quantifying the metabolic activity of lesions. However, due to the extremely thin walls of coronary arteries and their continuous motion throughout the cardiac cycle, the uptake of radioactive tracers is often significantly interfered with by background noise from the adjacent blood pool and some volume effects. Existing research indicates that simply measuring the SUV value at the vessel site is insufficient to accurately reflect the pathological state. Calculating the Target-to-Background Ratio (TBR) or performing background correction (e.g., subtracting the average SUV value from the atrial or ventricular blood pools) can significantly improve the specificity and accuracy of diagnosis.

[0004] Although quantitative analysis has significant clinical value, current image post-processing for coronary PET / CT mainly relies on manual operations by radiologists on imaging workstations. Physicians need to manually scan CT images layer by layer, identify the anatomical locations of the left anterior descending artery, circumflex artery, and right coronary artery, and manually delineate regions of interest (ROIs) to measure the maximum SUV (maximum value of the plaque). They also need to manually select background regions at the atrial or ventricular level. This traditional manual processing workflow has significant drawbacks: First, the course of coronary arteries is tortuous and complex, making manual delineation time-consuming and laborious, typically requiring 20 to 30 minutes to process a single case, which is insufficient for large-scale clinical screening. Second, manually selecting background locations is highly subjective, and measurement results from different operators exhibit significant inter-observer variability. If the background ROI is mistakenly selected within highly uptaken surrounding tissues, the corrected SUV value will be severely biased, directly affecting the assessment of plaque characteristics. In addition, most existing semi-automatic commercial software is based on simple threshold segmentation or region growing algorithms based on CT values. When the contrast agent filling is uneven or there are severe calcification artifacts, it often fails to accurately distinguish blood vessels from adjacent myocardial tissue, resulting in segmentation failure.

[0005] Therefore, there is an urgent need for a device that can automatically and accurately detect the background of blood vessels and atria, and realize automated SUV difference calculation and visualization to solve the problems of low efficiency, poor repeatability and lack of interactive experience of existing technologies. Summary of the Invention

[0006] This invention provides a method, apparatus, device, and storage medium for calculating cardiac coronary metabolic activity, with the main purpose of solving the problems of low efficiency, poor repeatability, and lack of interactive experience in existing technologies.

[0007] In a first aspect, embodiments of the present invention provide a method for calculating cardiac coronary metabolic activity, comprising:

[0008] S1. Input the target CT image data into the trained coronary artery key layer screening model. Use a binary classification mechanism to remove slices without vascular background and obtain key slices containing vascular features.

[0009] S2, For the key slice, digital image processing technology is used to achieve precise localization of the right atrium and three vessels through morphological operations, logical subtraction and PET / CT fusion analysis;

[0010] S3. For the right atrium and the three vessels, the right atrium automatic localization algorithm and SUV quantitative correction model are used to calculate the right atrium background SUV value and the maximum SUV value in the three vessel region by the target background difference, and the corrected plaque metabolism index is further output. The three vessels include the left anterior descending artery, the circumflex artery and the right coronary artery.

[0011] Furthermore, the procedure prior to step S1 includes:

[0012] Collect previous coronary artery PET and CT image samples from hospitals to construct a standard dataset containing positive and negative samples;

[0013] The sample data in the standard dataset are standardized by performing numerical correction, window width and window level adjustment and normalization in sequence to obtain the standardized standard dataset.

[0014] For the sample data in the standardized dataset, a spatiotemporal tensor is constructed using neighboring samples to obtain the enhanced standard dataset.

[0015] The coronary artery key layer screening model was trained using an enhanced standard dataset. During the training phase, random cropping, horizontal flipping, and angular rotation were performed to conduct binary classification training that included and excluded coronary arteries, resulting in a well-trained coronary artery key layer screening model.

[0016] Furthermore, the coronary artery key layer screening model is obtained by fusing a CNN model and a Transformer model. The coronary artery key layer screening model includes a feature extraction layer, a sequence modeling layer, a classification head, and a fully connected layer, wherein:

[0017] The feature extraction layer is used to select EfficientNet as the backbone network, load pre-trained weights and freeze shallow parameters, and only fine-tune deep parameters, so as to extract the spatial texture features of the target CT image data.

[0018] The sequence modeling layer is used to introduce sinusoidal position coding to mark the spatial order of the target CT image data. With the help of the coding layer of the Transformer model containing a multi-head self-attention mechanism, the anatomical continuity features of the target CT image data are captured.

[0019] The classification head is used to extract the feature vector of the sequence center position;

[0020] The fully connected layer is used to output classification probabilities based on the spatial texture features, the anatomical continuity features, and the feature vector of the sequence center position.

[0021] Furthermore, the training loss function adopts BCEWithLogitsLoss, and the pos_weight parameter is set to address the imbalance between positive and negative samples, forcing the coronary artery key layer screening model to focus on a minority of vascular samples.

[0022] The optimizer uses AdamW for parameter updates and is paired with a cosine annealing learning rate scheduler to dynamically adjust the learning rate, thus preventing the coronary artery key layer screening model from getting trapped in local optima.

[0023] Further, step S2 includes:

[0024] S21, For the key slice, the whole heart mask and the ventricular blood pool mask are obtained by using grayscale threshold truncation, morphological opening and closing operations and connected component analysis algorithms.

[0025] S22, Based on the whole heart mask and the ventricular blood pool mask, a candidate region including the coronary artery and its wall is obtained.

[0026] Further, step S3 includes:

[0027] S31, For the candidate region, the registration and fusion display technology of PET and CT images is used to obtain the fused image of the candidate region after registration;

[0028] S32, the fused image is converted to the HSV color space, and blood vessel pixels are selected in the candidate area by using a preset hue threshold and a preset saturation threshold;

[0029] S33, Based on the blood vessel pixels, identify the right atrial region on the target CT image data, and generate a statistical window at the corresponding position in the target PET image data to calculate the right atrial background SUV value;

[0030] S34. Establish a quantitative model based on the difference between the target and the background, calculate the maximum SUV value in the three-vessel region, subtract the background SUV value of the right atrium, and output the corrected plaque metabolic index.

[0031] Furthermore, step S4 is followed by:

[0032] The target CT image data, key slices, whole heart mask, ventricular blood pool mask, candidate region, fused image, vascular pixels, right atrial region and three vessel regions are displayed through a lightweight visualization device with B / S architecture.

[0033] In a second aspect, embodiments of the present invention provide a cardiac coronary metabolic activity calculation device, comprising:

[0034] The extraction module is used to input the target CT image data into the trained coronary artery key layer screening model, and remove non-vascular background slices through a binary classification mechanism to obtain key slices containing vascular features.

[0035] The positioning module is used to accurately locate the right atrium and three vessels in the key slices by using digital image processing technology, through morphological operations, logical subtraction and PET / CT fusion analysis.

[0036] The calculation module is used to calculate the baseline SUV value of the right atrium and the maximum SUV value within the three vessel regions by using the right atrium automatic localization algorithm and SUV quantitative correction model, and further output the corrected plaque metabolism index for the right atrium and the three vessels. The three vessels include the left anterior descending artery, the circumflex artery and the right coronary artery.

[0037] Thirdly, embodiments of the present invention provide a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described method for calculating cardiac coronary metabolic activity.

[0038] Fourthly, embodiments of the present invention provide a computer storage medium storing a computer program, which, when executed by a processor, implements the steps of the above-described method for calculating cardiac coronary metabolic activity.

[0039] This invention proposes a method, device, equipment, and storage medium for calculating cardiac coronary metabolic activity. It constructs a key coronary artery layer screening model that integrates a CNN model and a Transformer model, combining the feature extraction advantages of both models. It quickly eliminates avascular background slices through a binary classification mechanism, reducing computational load. At the same time, based on traditional digital image processing technology, it achieves cardiac substructure segmentation and precise vessel localization through morphological operations, logical subtraction, and PET / CT fusion analysis.

[0040] To address the challenges in plaque activity assessment, this invention will research an automatic right atrial localization algorithm and an SUV quantitative correction model. By calculating the target-background difference, the corrected plaque metabolic index will be output, providing an objective assessment indicator. In addition, a B / S architecture web-based visualization device will be developed to achieve high-performance rendering of DICOM images and will incorporate the right atrial selector function to support physician interactive fine-tuning, forming a closed-loop workflow of "algorithm recommendation + manual review". Attached Figure Description

[0041] Figure 1 A flowchart illustrating a method for calculating cardiac coronary metabolic activity provided in an embodiment of the present invention;

[0042] Figure 2 This is a schematic diagram of a cardiac coronary metabolic activity calculation device provided in an embodiment of the present invention.

[0043] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0044] The embodiments of this application are described in detail below. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain this application, and should not be construed as limiting this application.

[0045] To enable those skilled in the art to better understand the solutions of this application, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0046] In the embodiments of this application, "at least one" refers to one or more; "multiple" refers to two or more. In the description of this application, terms such as "first," "second," and "third" are used only for descriptive purposes and should not be construed as indicating or implying relative importance or order. Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0047] References such as “one embodiment” or “some embodiments” as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the terms “comprising,” “including,” “having,” and variations thereof, as used in this specification, mean “including, but not limited to,” unless otherwise specifically emphasized.

[0048] With the development of artificial intelligence technology, deep learning has made groundbreaking progress in the field of medical image analysis. Convolutional Neural Networks (CNNs) perform excellently in image feature extraction, but are limited by their local receptive fields, making it difficult to capture long-range dependencies in images. They are prone to breaks or misidentifications when processing long and continuous topological structures like coronary arteries. In recent years, the Transformer model has introduced a self-attention mechanism, which can effectively establish global contextual relationships within an image.

[0049] However, few existing publicly available technologies can organically combine the local detail extraction capabilities of CNN models with the global modeling capabilities of Transformer models and apply them to the complex logical analysis of PET / CT dual-modal data. In particular, existing technologies lack an integrated device that can automatically integrate CT anatomical localization, morphological background subtraction, PET metabolic quantification, and intuitive visualization on a web platform. Current research and clinical workflows are often fragmented, with data calculation and result presentation separated, lacking intuitive interactive methods to assist physicians in verifying and fine-tuning the automatically calculated background data.

[0050] Therefore, embodiments of the present invention provide a method for calculating cardiac coronary metabolic activity. Figure 1 A flowchart of a method for calculating cardiac coronary metabolic activity provided in an embodiment of the present invention is shown below. Figure 1 As shown, the method includes:

[0051] Among them, the SUV value, as a core parameter reflecting tissue metabolic activity in PET imaging data, is of great significance for clinical disease assessment due to its accurate extraction. This objective aims to construct a precise registration mechanism between PET and CT images based on the annotated atrial region in CT imaging data. This mechanism automatically matches the anatomical location corresponding to the atrium in the PET image and, through a standardized parameter extraction algorithm, accurately obtains the standard uptake value (SUV) of that atrial region. This completely eliminates the cumbersome process of manually matching images and measuring parameters, effectively avoiding measurement errors caused by manual operation, and significantly improving the efficiency and accuracy of SUV value extraction. This provides reliable basic data support for subsequent calculation of vascular metabolic differentials.

[0052] S1. Input the target CT image data into the trained coronary artery key layer screening model. Use a binary classification mechanism to remove slices without vascular background and obtain key slices containing vascular features.

[0053] To address the issues of large data volume and small proportion of vascular layers in PET / CT tomographic image sequences, this invention will study and construct an efficient "coarse screening" classification model for key coronary artery layers based on the fusion of CNN and Transformer. The focus is on how to combine the local feature extraction capability of the CNN model with the global attention mechanism of the Transformer model. The CNN module is used to extract texture and edge features from a single CT image, while the Transformer module is used to capture the anatomical context information of the CT image data.

[0054] Meanwhile, the trained coronary artery key layer screening model performs binary classification on the input target CT image data, classifying it as "containing coronary arteries / not containing coronary arteries". This coronary artery key layer screening model quickly removes a large number of background slices without vessels, and only the key slices containing vascular features are subjected to subsequent fine processing, thereby significantly reducing the computational load of the present invention and reducing false positives.

[0055] S2, For the key slice, digital image processing technology is used to achieve precise localization of the right atrium and three vessels through morphological operations, logical subtraction and PET / CT fusion analysis;

[0056] Precise localization of the three coronary arteries is a key prerequisite for subsequent metabolic parameter analysis. Traditional manual localization methods not only require operators to have extensive clinical imaging experience, but also involve complex and time-consuming procedures. Furthermore, the localization results are easily affected by personal experience, resulting in poor consistency.

[0057] This invention, based on the key slices obtained in the previous step, combines the anatomical morphology, course patterns, and CT image density differences of the three coronary arteries to construct a dedicated automated identification and localization model. Through algorithm optimization, it achieves accurate differentiation and individual localization of the three vessels, clearly defining the specific anatomical range of each vessel. This effectively solves the core pain points of traditional vessel localization, such as complex manual operation, long processing time, and high subjectivity, ensuring the accuracy of the three vessel localization and the consistency of localization in different scenarios. It also provides accurate localization basis for the extraction of metabolic parameters of the corresponding vessel region in subsequent PET images.

[0058] S3. For the right atrium and the three vessels, the right atrium automatic localization algorithm and SUV quantitative correction model are used to calculate the right atrium background SUV value and the maximum SUV value in the three vessel region by the target background difference, and the corrected plaque metabolism index is further output. The three vessels include the left anterior descending artery, the circumflex artery and the right coronary artery.

[0059] The vascular metabolic difference (i.e., the maximum SUV value of each vessel minus the atrial SUV value) is an important indicator for assessing abnormal coronary artery vascular metabolism. Traditional calculation methods require manual extraction of the atrial SUV value and the maximum SUV value of each vessel, and then manual calculation of the difference. The process is cumbersome and prone to calculation errors, making it difficult to meet the needs of efficient clinical diagnosis and treatment.

[0060] This invention automatically matches the corresponding vascular region in PET images based on the calibrated locations of three blood vessels using image registration technology. It employs a precise parameter extraction algorithm to extract the maximum SUV value of each vascular region and simultaneously calls the automatically extracted atrial SUV value. Through preset standardized calculation logic, it automatically calculates the difference between the maximum SUV value of each blood vessel and the atrial SUV value, automatically generating standardized and accurate calculation results. No manual intervention is required in any calculation process, reducing human calculation errors, improving the standardization and efficiency of data processing, and providing clinicians with reliable vascular metabolic assessment data.

[0061] In some embodiments, the method further includes the following steps prior to step S1:

[0062] Collect previous coronary artery PET and CT image samples from hospitals to construct a standard dataset containing positive and negative samples;

[0063] The sample data in the standard dataset are standardized by performing numerical correction, window width and window level adjustment and normalization in sequence to obtain the standardized standard dataset.

[0064] For the sample data in the standardized dataset, a spatiotemporal tensor is constructed using neighboring samples to obtain the enhanced standard dataset.

[0065] The coronary artery key layer screening model was trained using an enhanced standard dataset. During the training phase, random cropping, horizontal flipping, and angular rotation were performed to conduct binary classification training that included and excluded coronary arteries, resulting in a well-trained coronary artery key layer screening model.

[0066] This invention first involves constructing a standard dataset source:

[0067] We collected previous coronary PET / CT scan data from hospitals and used these scan data as sample data, covering original DICOM format CT thin-slice images (i.e. CT image samples) and corresponding PET attenuation-corrected images (i.e. PET image samples). This enabled us to establish a standard dataset containing "positive samples (layers containing coronary arteries and their calcifications / plaques)" and "negative samples (cardiac background layers excluding blood vessels)".

[0068] DICOM format is an international standard format for medical digital imaging and communication, widely used for storing, transmitting, and processing medical image data. It not only contains image pixel information but also integrates metadata such as patient identity, equipment parameters, and examination details, ensuring that image and medical record information are inseparable.

[0069] Then, DICOM data standardization is performed: using the pydicom library to parse DICOM metadata (i.e., sample data in the standard dataset), CT value correction is first completed. Specifically, the RescaleSlope and RescaleIntercept parameters are read to convert the original pixel values ​​into standard Hounsfield units (HU).

[0070] Next, the window width and window level are adjusted, and a soft tissue window is applied to perform truncation processing to enhance the contrast between blood vessels and surrounding fat. Finally, normalization processing is carried out to linearly map the pixel values ​​to the [0, 255] interval and convert them into a three-channel RGB image to adapt to the input requirements of the pre-trained coronary artery key layer screening model.

[0071] To address the issue of insufficient information in 2D slices, a 2.5D sequence construction and enhancement were further completed: the "2.5D slice stacking method" was adopted, taking the current slice as the center and taking two adjacent slices forward and backward to construct a spatiotemporal tensor with the shape (B, 5, 3, H, W).

[0072] Meanwhile, during the training phase of the coronary artery key layer screening model, online data augmentation techniques such as RandomResizedCrop, HorizontalFlip, and RandomRotation were introduced to improve the generalization ability of the coronary artery key layer screening model.

[0073] In some embodiments, the coronary artery key layer screening model is obtained by fusing a CNN model and a Transformer model. The coronary artery key layer screening model includes a feature extraction layer, a sequence modeling layer, a classification head, and a fully connected layer, wherein:

[0074] The feature extraction layer is used to select EfficientNet as the backbone network, load pre-trained weights and freeze shallow parameters, and only fine-tune deep parameters, so as to extract the spatial texture features of the target CT image data.

[0075] The sequence modeling layer is used to introduce sinusoidal position coding to mark the spatial order of the target CT image data. With the help of the coding layer of the Transformer model containing a multi-head self-attention mechanism, the anatomical continuity features of the target CT image data are captured.

[0076] The classification head is used to extract the feature vector of the sequence center position;

[0077] The fully connected layer is used to output classification probabilities based on the spatial texture features, the anatomical continuity features, and the feature vector of the sequence center position.

[0078] In this embodiment of the invention, a CNN-Transformer hybrid model is built based on the PyTorch deep learning framework. The overall architecture of the coronary artery key layer screening model consists of three parts: a feature extraction layer, a sequence modeling layer, and a classification head.

[0079] The feature extraction layer (Backbone) uses EfficientNet as the backbone network. After loading pre-trained weights, the shallow layer parameters are frozen and only the deep layer parameters are fine-tuned to extract the spatial texture features of each frame slice.

[0080] The sequence modeling layer (Neck) introduces sinusoidal positional encoding to mark the spatial order of slices, and uses the TransformerEncoderLayer, which includes a multi-head attention mechanism, to capture the anatomical continuity features between slices.

[0081] The classification head extracts the feature vector at the center of the sequence and outputs the classification probability through a fully connected layer (Linear).

[0082] In terms of training strategy optimization, the loss function adopted is BCEWithLogitsLoss, and the pos_weight parameter is set to address the problem of imbalanced positive and negative sample ratios, forcing the model to focus on the minority class of blood vessel samples. The optimizer uses AdamW for parameter updates, and is paired with CosineAnnealingLR (cosine annealing learning rate scheduler) to dynamically adjust the learning rate and avoid the model getting stuck in local optima. At the same time, an early stopping mechanism is introduced to monitor the accuracy and loss value of the validation set in real time. When the indicators no longer improve, the best model weights (best_model.pth) are automatically saved, and finally the trained coronary artery key layer screening model is obtained.

[0083] In some embodiments, step S2 includes:

[0084] S21, For the key slice, the whole heart mask and the ventricular blood pool mask are obtained by using grayscale threshold truncation, morphological opening and closing operations and connected component analysis algorithms.

[0085] S22, Based on the whole heart mask and the ventricular blood pool mask, a candidate region including the coronary artery and its wall is obtained.

[0086] In this embodiment of the invention, the anatomical structure separation algorithm is based on the OpenCV function library to construct a morphological processing flow:

[0087] First, the cv2.threshold function is called to perform binarization segmentation on the key slices. Then, the morphological operation of the cv2.morphologyEx function is used to fill the cavity regions inside the heart, generating a "whole heart mask" that covers the entire heart area.

[0088] Next, connected component analysis is performed using the cv2.connectedComponents function to filter out the largest bright connected region in the image, which is defined as the "ventricular blood pool mask".

[0089] Finally, the logical operation Vessel = Heart ∩ (¬Ventricle) is performed, and the opening morphological operation is applied to the result to remove noise interference, thus obtaining the candidate region of the coronary artery.

[0090] In some embodiments, step S3 includes:

[0091] S31, For the candidate region, the registration and fusion display technology of PET and CT images is used to obtain the fused image of the candidate region after registration;

[0092] S32, the fused image is converted to the HSV color space, and blood vessel pixels are selected in the candidate area by using a preset hue threshold and a preset saturation threshold;

[0093] S33, Based on the blood vessel pixels, identify the right atrial region on the target CT image data, and generate a statistical window at the corresponding position in the target PET image data to calculate the right atrial background SUV value;

[0094] S34. Establish a quantitative model based on the difference between the target and the background, calculate the maximum SUV value in the three-vessel region, subtract the background SUV value of the right atrium, and output the corrected plaque metabolic index.

[0095] In the multimodal fusion and extraction stage of this invention, spatial registration and resolution resampling are first performed on PET image data and CT image data to achieve spatial alignment of the two modal images. Then, the fused image after candidate region registration is converted from RGB color space to HSV color space. A preset hue threshold range (such as yellow to red band) and saturation threshold range are set. High metabolic pixels that meet the threshold conditions are screened and extracted within the above-mentioned coronary artery candidate region to complete the feature extraction of the target region.

[0096] The background automatic correction module develops an automatic right atrial localization algorithm. After identifying the anatomical region of the right atrium through this algorithm, it calculates the mean and standard deviation of the standard uptake values ​​(SUV) within that region.

[0097] Based on the above calculation results, a quantitative correction formula is constructed: TBR=SUVmax(vessel)-(SUVmean(atrium)+2×SUVstd(atrium)). This formula enables automated quantitative assessment of coronary artery metabolic levels, and background correction and numerical calculation can be completed without manual intervention.

[0098] It should also be noted that during clinical diagnosis and treatment, medical staff need to quickly and clearly obtain imaging localization information and metabolic parameter results. Traditional textual results presentation methods are difficult to intuitively reflect the correspondence between localization location and parameters, and are difficult to interpret.

[0099] This invention aims to integrate the core data and results generated throughout the entire process, including atrial positioning and labeling information, atrial SUV value, three vessel positioning and labeling information, maximum SUV value of each vessel and corresponding metabolic difference calculation results, and to design an intuitive, clear and clinically relevant automated visualization module.

[0100] This module needs to achieve the synchronous presentation of image localization annotation and metabolic parameter results, clearly marking the specific locations of the atria and three vessels in PET and CT images, and synchronously displaying the values ​​and differences of various parameters, so as to facilitate clinical medical staff to quickly view and interpret image data and analysis results, reduce the difficulty of result interpretation, and improve the convenience and efficiency of clinical applications.

[0101] Therefore, in some embodiments, step S4 is followed by:

[0102] The target CT image data, key slices, whole heart mask, ventricular blood pool mask, candidate region, fused image, vascular pixels, right atrial region and three vessel regions are displayed through a lightweight visualization device with B / S architecture.

[0103] To achieve seamless transformation from algorithm models to clinical applications, this invention will construct a lightweight, interactive PET / CT coronary metabolic assessment device based on a browser / server (B / S) architecture, focusing on solving the problems of difficult deployment, poor interactivity, and data silos in traditional desktop workstations.

[0104] In terms of backend service architecture and API encapsulation, the device uses Python as the core development language and utilizes the FastAPI or Flask framework to build high-performance RESTful API interfaces. The aforementioned "coronary artery key layer screening model integrating CNN-Transformer", "anatomical topology logical segmentation algorithm" and "HSV domain quantitative correction module" are encapsulated as independent microservice components. After the calculation is completed, the backend not only returns quantitative results in JSON format containing the maximum SUV of the three blood vessels and the mean / standard deviation of the right atrial background SUV, but also generates serialized image data containing automatically labeled masks to ensure that the frontend can accurately reproduce the algorithm's localization results.

[0105] In terms of front-end visualization engine and DICOM rendering, it is developed based on Vue.js or React framework, and integrates Cornerstone.js or OHIF open source medical imaging library to achieve native rendering of DICOM images on the browser side; by developing a custom layer blending algorithm, it supports real-time overlay of CT grayscale images and PET pseudo-color heatmaps, and uses HTML5Canvas or WebGL technology to realize dynamic adjustment of window width and window level, image scaling, translation, and stepless sliding adjustment of PET / CT fusion transparency, so that doctors can clearly observe the correspondence between high metabolic lesions and anatomical structures.

[0106] In summary, this invention adopts a technical approach that deeply integrates various techniques, including "deep sequence learning screening, anatomical topological logical segmentation, bimodal color domain quantization, and human-computer collaborative interaction." Addressing the pain points in coronary artery vulnerable plaque assessment, such as small vessel size, significant motion artifact interference, and strong subjectivity in baseline calculation, it is implemented through the following key technical steps:

[0107] First, the patient's raw PET / CT sequences (i.e., PET and CT image data) are read using a DICOM protocol parsing interface (such as pydicom). Key metadata (such as rescale slope / intercept, patient weight, injection dosage, etc.) is parsed, and the raw CT data is mapped to standard Hounsfield units (HU). Soft tissue windowing is then applied to enhance the contrast between blood vessels and surrounding fat. Building upon this, this invention overcomes the limitations of traditional single-frame image processing by constructing a "2.5D spatiotemporal sequence tensor." This involves taking the current slice to be detected as the center and extracting N adjacent slices (e.g., 2 slices before and after, for a total of 5 slices) along the Z-axis as context slices. After normalization and size resampling, an input tensor containing inter-slice continuity information is generated, providing rich spatial context features for subsequent models and completing the standardized construction of multimodal data based on the 2.5D spatiotemporal tensor.

[0108] To address the challenge of small and discontinuous coronary arteries, this invention employs a hybrid network architecture combining local features and global attention to perform hierarchical screening of the input 2.5D tensor. First, a pre-trained EfficientNet is used as the feature extraction backbone to extract high-dimensional texture and edge features from each frame of the sequence in parallel. Then, sinusoidal positional encoding is introduced to mark the spatial relative positions of the slices. The feature sequence is then input into a Transformer encoder, utilizing multi-head self-attention to capture the anatomical continuity between slices. The model ultimately outputs the confidence probability of the central slice. This key technology enables the device to refer to the anatomical features of adjacent layers when determining whether the current layer contains a blood vessel, effectively solving the problem of misjudgment of vessel rupture or noise caused by partial volume effects. This achieves highly sensitive automatic screening, completing intelligent coronary artery screening based on a CNN-Transformer hybrid architecture.

[0109] Based on the selected positive layers, this invention employs morphological segmentation technology based on anatomical priors, abandoning expensive pixel-level supervised learning and instead utilizing the significant differences in CT values ​​and geometric connectivity between the heart and ventricles. First, a "whole heart contour mask" is constructed through threshold segmentation and morphological closing operations. Second, connected component analysis is used to extract the high-brightness "ventricular blood pool mask." Finally, a logical subtraction operation (Target = Heart ∩ ¬Ventricle) is performed, supplemented by opening operations to remove noise. This key technology cleverly utilizes anatomical topological relationships to remove the high-brightness ventricular blood pool interference from the overall heart, thereby accurately locating the anatomical region containing the coronary arteries and their walls without manual annotation, completing unsupervised vascular region extraction based on anatomical topological logic.

[0110] To accurately quantify plaque activity, this invention proposes a fusion analysis technique based on the HSV color space. The located vascular region is mapped to the PET image space, and the fused image is converted to the HSV (hue, saturation, brightness) color space. Utilizing the unique hue distribution (e.g., yellow to red range) and saturation threshold of inflammatory lesions in the PET thermal image, an adaptive filter is constructed to accurately locate high-metabolic pixels and calculate the maximum SUV value. Simultaneously, the geometric center of the right atrium is automatically located, and a background statistical window is established to calculate the background correction value. The final index is output using the formula SUVcorrected = SUVmax - SUVbackground. This technique effectively filters low-metabolic background noise, ensuring the objectivity of the quantification results and completing adaptive fusion and quantitative correction in the HSV domain of PET / CT dual-modality imaging.

[0111] To address the uncertainties of fully automated algorithms in complex pathological conditions, this invention constructs a lightweight interactive device based on a B / S architecture. The front-end uses HTML5 / WebGL technology to replicate the image reading experience of a desktop workstation, while the back-end encapsulates image processing logic as an API interface. This not only renders and merges images and quantification results in real time but also introduces a "human-in-the-loop" mechanism, allowing doctors to interact via mouse clicks on the web interface to reset the right atrial baseline sampling center or fine-tune the vascular ROI in real time. The back-end recalculates the TBR or difference based on the corrected coordinates in real time, preserving the high efficiency of automated algorithms while ensuring high reliability and security for clinical diagnosis through a manual verification mechanism, thus completing a web-based "human-in-the-loop" interactive visualization.

[0112] This invention innovatively proposes a coronary artery intelligent assessment and interaction device based on a "2.5D spatiotemporal sequence context." First, a CNN-Transformer hybrid architecture is constructed, utilizing a global attention mechanism across multiple slice sequences to solve the problem of difficulty in identifying small vessels due to motion artifacts. Second, a pioneering anatomical topological logic segmentation algorithm based on "whole heart minus ventricular blood pool" achieves unsupervised vascular region extraction without pixel-level manual annotation. Simultaneously, PET / CT dual-modal HSV color gamut adaptive analysis technology is introduced to accurately locate high-metabolic lesions and automatically subtract the right atrial background, establishing an objective SUV difference quantification standard. Finally, a closed-loop diagnosis and treatment process of "automatic algorithm recommendation + real-time manual verification" is implemented based on a Web architecture, significantly improving the automation level and clinical reliability of vulnerable plaque assessment.

[0113] Figure 2 This is a schematic diagram of a cardiac coronary metabolic activity calculation device provided in an embodiment of the present invention, as shown below. Figure 2 As shown, the device includes:

[0114] The extraction module 210 is used to input the target CT image data into the trained coronary artery key layer screening model, and remove non-vascular background slices through a binary classification mechanism to obtain key slices containing vascular features.

[0115] The positioning module 220 is used to accurately locate the right atrium and three vessels for the key slice by using digital image processing technology, through morphological operations, logical subtraction and PET / CT fusion analysis.

[0116] The calculation module 230 is used to calculate the background SUV value of the right atrium and the maximum SUV value in the three-vessel region by using the right atrium automatic localization algorithm and SUV quantitative correction model for the right atrium and the three-vessel region, and further output the corrected plaque metabolism index. The three-vessel region includes the left anterior descending artery, the circumflex artery and the right coronary artery.

[0117] This embodiment is a device embodiment corresponding to the above method embodiment. Its specific process is the same as that of the above method embodiment. For details, please refer to the above method embodiment. This device embodiment will not repeat the details.

[0118] Each module in the aforementioned cardiac coronary metabolic activity calculation device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0119] This invention also provides a computer device, which may be a server. The computer device includes a processor, a memory, a network interface, and a database connected via a device bus. The processor provides computational and control capabilities. The memory includes a computer storage medium and internal memory. The computer storage medium stores operating devices, computer programs, and a database. The internal memory provides an environment for the operation of the operating devices and computer programs in the computer storage medium. The database stores data generated or acquired during the execution of a method for calculating coronary artery metabolic activity, such as key slices and target CT image data. The network interface communicates with external terminals via a network connection. When the computer program is executed by the processor, it implements a method for calculating coronary artery metabolic activity.

[0120] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of a method for calculating cardiac coronary metabolic activity as described in the above embodiment. Alternatively, when the processor executes the computer program, it implements the functions of each module / unit in the embodiment of a device for calculating cardiac coronary metabolic activity.

[0121] In one embodiment, a computer storage medium is provided, on which a computer program is stored. When executed by a processor, the computer program implements the steps of the method for calculating cardiac coronary metabolic activity described in the above embodiment. Alternatively, when executed by a processor, the computer program implements the functions of each module / unit in the above embodiment of the apparatus for calculating cardiac coronary metabolic activity.

[0122] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0123] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0124] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method for calculating cardiac coronary metabolic activity, characterized in that, include: S1. Input the target CT image data into the trained coronary artery key layer screening model. Use a binary classification mechanism to remove slices without vascular background and obtain key slices containing vascular features. S2, For the key slice, digital image processing technology is used to achieve precise localization of the right atrium and three vessels through morphological operations, logical subtraction and PET / CT fusion analysis; S3. For the right atrium and the three vessels, the right atrium automatic localization algorithm and SUV quantitative correction model are used to calculate the right atrium background SUV value and the maximum SUV value in the three vessel region by the target background difference, and the corrected plaque metabolism index is further output. The three vessels include the left anterior descending artery, the circumflex artery and the right coronary artery.

2. The method for calculating cardiac coronary metabolic activity according to claim 1, characterized in that, Step S1 is preceded by: Collect previous coronary artery PET and CT image samples from hospitals to construct a standard dataset containing positive and negative samples; The sample data in the standard dataset are standardized by performing numerical correction, window width and window level adjustment and normalization in sequence to obtain the standardized standard dataset. For the sample data in the standardized dataset, a spatiotemporal tensor is constructed using neighboring samples to obtain the enhanced standard dataset. The coronary artery key layer screening model was trained using an enhanced standard dataset. During the training phase, random cropping, horizontal flipping, and angular rotation were performed to conduct binary classification training that included and excluded coronary arteries, resulting in a well-trained coronary artery key layer screening model.

3. The method for calculating cardiac coronary metabolic activity according to claim 1, characterized in that, The coronary artery key layer screening model is obtained by fusing a CNN model and a Transformer model. The coronary artery key layer screening model includes a feature extraction layer, a sequence modeling layer, a classification head, and a fully connected layer, wherein: The feature extraction layer is used to select EfficientNet as the backbone network, load pre-trained weights and freeze shallow parameters, and only fine-tune deep parameters, so as to extract the spatial texture features of the target CT image data. The sequence modeling layer is used to introduce sinusoidal position coding to mark the spatial order of the target CT image data. With the help of the coding layer of the Transformer model containing a multi-head self-attention mechanism, the anatomical continuity features of the target CT image data are captured. The classification head is used to extract the feature vector of the sequence center position; The fully connected layer is used to output classification probabilities based on the spatial texture features, the anatomical continuity features, and the feature vector of the sequence center position.

4. The method for calculating cardiac coronary metabolic activity according to claim 2, characterized in that, The training loss function uses BCEWithLogitsLoss, and the pos_weight parameter is set to address the imbalance between positive and negative samples, forcing the coronary artery key layer screening model to focus on a minority of vascular samples. The optimizer uses AdamW for parameter updates and is paired with a cosine annealing learning rate scheduler to dynamically adjust the learning rate, thus preventing the coronary artery key layer screening model from getting trapped in local optima.

5. The method for calculating cardiac coronary metabolic activity according to claim 1, characterized in that, Step S2 includes: S21, For the key slice, the whole heart mask and the ventricular blood pool mask are obtained by using grayscale threshold truncation, morphological opening and closing operations and connected component analysis algorithms. S22, Based on the whole heart mask and the ventricular blood pool mask, a candidate region including the coronary artery and its wall is obtained.

6. The method for calculating cardiac coronary metabolic activity according to claim 1, characterized in that, Step S3 includes: S31, For the candidate region, the registration and fusion display technology of PET and CT images is used to obtain the fused image of the candidate region after registration; S32, the fused image is converted to the HSV color space, and blood vessel pixels are selected in the candidate area by using a preset hue threshold and a preset saturation threshold; S33, Based on the blood vessel pixels, identify the right atrial region on the target CT image data, and generate a statistical window at the corresponding position in the target PET image data to calculate the right atrial background SUV value; S34. Establish a quantitative model based on the difference between the target and the background, calculate the maximum SUV value in the three-vessel region, subtract the background SUV value of the right atrium, and output the corrected plaque metabolic index.

7. The method for calculating cardiac coronary metabolic activity according to any one of claims 1 to 6, characterized in that, Step S4 is followed by: The target CT image data, key slices, whole heart mask, ventricular blood pool mask, candidate region, fused image, vascular pixels, right atrial region and three vessel regions are displayed through a lightweight visualization device with B / S architecture.

8. A device for calculating cardiac coronary metabolic activity, characterized in that, include: The extraction module is used to input the target CT image data into the trained coronary artery key layer screening model, and remove non-vascular background slices through a binary classification mechanism to obtain key slices containing vascular features. The positioning module is used to accurately locate the right atrium and three vessels in the key slices by using digital image processing technology, through morphological operations, logical subtraction and PET / CT fusion analysis. The calculation module is used to calculate the baseline SUV value of the right atrium and the maximum SUV value within the three vessel regions by using the right atrium automatic localization algorithm and SUV quantitative correction model, and further output the corrected plaque metabolism index for the right atrium and the three vessels. The three vessels include the left anterior descending artery, the circumflex artery and the right coronary artery.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method for calculating cardiac coronary metabolic activity as described in any one of claims 1 to 7.

10. A computer storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the method for calculating cardiac coronary metabolic activity as described in any one of claims 1 to 7.

Images