STEMI patient two-year MACE event prediction method and system based on multi-sequence CMR

Abstract

The invention relates to the field of cardiac magnetic resonance image analysis, and discloses an STEMI patient two-year MACE event prediction method based on multi-sequence CMR, and the method comprises the following steps: obtaining a plurality of CMR sequence images and clinical data of an acute stage of an STEMI patient; segmenting and extracting the CMR sequence image by using a heart ROI extraction model to obtain a heart region CMR image, and preprocessing the heart region image; establishing a prediction model by using a heart area CMR image and clinical data, randomly initializing model parameters, and performing model training by using a training set sample; and using the trained prediction model to predict the two-year MACE event occurrence probability of the patient. According to the two-year MACE event prediction model, a plurality of CMR sequences and clinical data are adopted to predict the two-year MACE event, and meanwhile, the reliability of a prediction result is improved by using a multi-sequence image and multi-modal data.

Description

Technical Field

[0001] This invention relates to the field of cardiac magnetic resonance imaging analysis, and more particularly to a method and system for predicting two-year MACE events in STEMI patients based on multi-sequence CMR. Background Technology

[0002] Acute myocardial infarction is one of the leading causes of morbidity and mortality worldwide. Even with timely reperfusion therapy, patients with ST-segment elevation myocardial infarction (STEMI) remain at high risk of major adverse cardiovascular events (MACE). Identifying high-risk STEMI patients is crucial for guiding optimized and personalized precision treatment in clinical settings.

[0003] Therefore, considerable effort has been devoted to developing tools for prognostic risk stratification in STEMI patients, including clinical scores, biomarkers, and imaging methods. However, most clinical risk scores were developed before the reperfusion era and may not fully explain the impact of contemporary reperfusion therapies on patient outcomes. Furthermore, traditional imaging parameters used in routine clinical practice, such as left ventricular ejection fraction (LVEF) measured by echocardiography, while prognostic, fail to reflect the severity of myocardial pathological tissue damage.

[0004] Therefore, there is an urgent need for a more effective method for predicting MACE events. Summary of the Invention

[0005] The main objective of this invention is to address the problems existing in the prediction of MACE events in the prior art. A two-year MACE event prediction method for STEMI patients based on multi-sequence CMR includes the following steps: A two-year MACE event prediction method for STEMI patients based on multi-sequence CMR is characterized by the following steps: Acquire multiple CMR sequence images and clinical data from STEMI patients during the acute phase; The CMR sequence images were segmented and extracted using a cardiac ROI extraction model to obtain cardiac region CMR images, which were then preprocessed. A predictive model was built using cardiac region CMR images and clinical data. The model parameters were randomly initialized, and the model was trained using training set samples. Use a trained predictive model to predict the probability of MACE events occurring in patients over two years.

[0006] A second aspect of the present invention provides a two-year MACE event prediction system for STEMI patients based on multi-sequence CMR, comprising: The data acquisition unit is used to acquire multiple CMR sequence images and clinical data of STEMI patients in the acute phase. The preprocessing unit is used to segment and extract the CMR sequence image using a cardiac ROI extraction model to obtain a cardiac region CMR image and preprocess the cardiac region image. The model building and training unit is used to build a predictive model using cardiac region CMR images and clinical data, randomly initialize model parameters, and train the model using training set samples. The prediction unit is used to predict the probability of a patient's MACE event occurring over two years using a trained prediction model.

[0007] A third aspect of the present invention provides an electronic device comprising: a memory and at least one processor, the memory storing instructions, the memory and the at least one processor being interconnected via a circuit; the at least one processor calling the instructions in the memory to cause the electronic device to perform the above-described method for predicting two-year MACE events in STEMI patients based on multi-sequence CMR.

[0008] A fourth aspect of the invention provides a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the above-described method for predicting two-year MACE events in STEMI patients based on multi-sequence CMR.

[0009] The present invention has the following beneficial effects: This invention proposes a two-year method, DeepSTEMI, for predicting major adverse cardiovascular events (MACE) in patients with ST-segment elevation myocardial infarction based on multi-sequence magnetic resonance imaging. DeepSTEMI consists of a cardiac ROI extraction model and a two-year MACE event prediction model.

[0010] The cardiac ROI extraction model of the present invention is used to extract the cardiac region from each original CMR image.

[0011] The predictive model of this invention uses multiple CMR sequences (Cine, LGE, and T2-STIR) and clinical data (age, sex, Grace score, and Killip score) to predict MACE events over two years. At the same time, the use of multi-sequence imaging and multimodal data increases the reliability of the prediction results. Attached Figure Description

[0012] Figure 1 This is a flowchart of the present invention.

[0013] Figure 2 This is an example diagram showing the prediction results of the present invention. Detailed Implementation

[0014] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” or “having,” and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0015] This invention addresses the problems existing in the two-year MACE event prediction methods for ST-segment elevation myocardial infarction (STEMI) patients. The invention aims to develop a deep learning-based method, DeepSTEMI, for predicting two-year MACE events in STEMI patients. The deep learning method of this invention consists of a cardiac ROI extraction model and a two-year MACE event prediction model. The cardiac ROI extraction model uses a UNet network to extract cardiac regions from each raw CMR image. The Transformer-based two-year MACE event prediction model uses multiple CMR sequences (Cine, LGE, and T2-STIR) and clinical data (age, sex, Grace score, and Killip score) to predict two-year MACE events.

[0016] Cine MRI Cine MRI is a dynamic, cinematic imaging sequence that uses steady-state free precession or gradient echo techniques. A series of images are rapidly acquired at different points in time during the cardiac cycle (systole and diastole) (typically 20-30 frames per heartbeat), and then these images are played back sequentially to create a cinematic effect similar to a heartbeat.

[0017] LGE (Late Gadolinium Enhancement) LGE is a T1-weighted imaging sequence performed after a period of time (usually 10–20 minutes) following an intravenous injection of gadolinium contrast agent. It uses specific techniques (such as phase-sensitive inversion recovery) to suppress signals from normal myocardium, making normal myocardium appear dark gray or black.

[0018] Light contrast imaging (LGE) is one of the "gold standards" for diagnosing myocardial infarction. The infarcted area (necrotic tissue replaced by fibrous scar tissue) will show significant enhancement (bright white) because the contrast agent remains in the fibrotic / necrotic tissue for a longer time, has a larger distribution volume, and is cleared more slowly. LGE can precisely locate the infarct site (which coronary artery lesion corresponds to), assess the infarct extent (transmural extent: subendocardial, transmural), and size.

[0019] T2-STIR (T2-weighted Short Tau Inversion Recovery) T2-STIR is a T2-weighted imaging sequence that incorporates STIR (Short Time Inversion Recovery) technology. STIR technology can effectively suppress high signal intensity in adipose tissue while enhancing the signal intensity of free water (such as edema).

[0020] In the early stages of an acute myocardial infarction (usually within days to weeks), ischemic injury leads to myocardial cell edema, and the affected area appears as a high signal (bright) on T2-STIR. This helps determine whether the myocardial injury is acute (edema present) or old (edema subsiding). Combining LGE (bright) and T2-STIR (bright) signals can differentiate between the infarct core (bright LGE, T2-STIR may be bright or not) and the surrounding dangerous myocardium (bright T2-STIR, but not bright LGE - indicating salvageable myocardium).

[0021] For ease of understanding, the specific process of the embodiments of the present invention is described below. Please refer to [link / reference]. Figure 1-2 The first embodiment of the method for predicting two-year MACE events in STEMI patients based on multi-sequence CMR in this invention includes: Acquire multiple CMR sequence images and clinical data from STEMI patients during the acute phase; The CMR sequence images were segmented and extracted using a cardiac ROI extraction model to obtain cardiac region CMR images, which were then preprocessed. A predictive model was built using cardiac region CMR images and clinical data. The model parameters were randomly initialized, and the model was trained using training set samples. Use a trained predictive model to predict the probability of MACE events occurring in patients over two years.

[0022] Preferably, the second step of the above process involves: considering that the heart region is only distributed in a local area of ​​the CMR image, it is necessary to first segment and crop the heart region on the image before performing MACE event prediction. The heart ROI extraction model uses a U-Net network for heart region segmentation and extraction. The U-Net model is an encoder-decoder architecture for image segmentation, employing a symmetrical U-shaped design. The encoder extracts features progressively through convolutional and downsampling layers, while the decoder reconstructs spatial details through upsampling and skip connections. These skip connections directly fuse high-resolution encoder features with the decoder output. Its fully convolutional nature allows it to handle input images of arbitrary size without resizing, while skip connections bridge the semantic and spatial information in the network depth, thus accurately locating boundaries. The heart ROI extraction model was pre-trained on the public ACDC dataset and fine-tuned on a small-scale multi-CMR sequence dataset containing heart region annotations collected in this study. The model was trained for 50 epochs in both the pre-training and fine-tuning steps, with a batch size of 32. Using the trained heart ROI extraction model, CMR images can be automatically cropped based on the predicted heart region. All models are implemented using the PyTorch framework.

[0023] Preferably, the third step above specifically involves a two-year MACE event prediction model based on multi-sequence magnetic resonance imaging. This step uses multiple CMR sequences extracted from the cardiac region and some clinical data as input to predict two-year MACE events. The clinical data is embedded using a multilayer perceptron (MLP) to generate a unified feature vector. VideoSwin Transformer and Swin Transformer are selected to extract features from motion (cinema) and static (LGE and T2-STIR) CMR sequences, respectively. Swin Transformer is a hierarchical visual transformer that introduces window-based multi-head self-attention (W-MSA) and shift-window multi-head self-attention (SW-MSA) mechanisms to effectively model local and global dependencies.

[0024] Preferably, the W-MSA and SW-MSA of the present invention are based on an attention mechanism, and their calculation formula is as follows:

[0025] The Swin Transformer and Video Swin Transformer networks alternately use window-based multi-head self-attention (W-MSA) and shift-window multi-head self-attention (SW-MSA) mechanisms. Two consecutive Swin Transformer blocks are computed as follows:

[0026]

[0027]

[0028]

[0029] By dividing the image into non-overlapping windows and computing a self-attention mechanism within each window, the computational complexity relative to the input size is reduced from quadratic to linear. The shifting window mechanism in alternating layers supports cross-window communication while maintaining computational efficiency, eliminating the need for a global attention mechanism. The Video Swin Transformer is a spatiotemporal extension of the Swin Transformer, adapting the hierarchical window attention mechanism from 2D images to 3D motion CMR data. Each CMR sequence is embedded into a uniform feature vector of dimension 256. All extracted features are further summed for two-year MACE event prediction. The entire two-year MACE event prediction model is trained end-to-end using a weighted cross-entropy loss function for 100 epochs. The batch size is set to 8. All models are implemented using the PyTorch framework.

[0030] The effectiveness of this invention is further illustrated by the following experiments. This invention verifies the proposed method for predicting major adverse cardiovascular events over two years from both qualitative and quantitative perspectives.

[0031] Qualitative experimental results This invention was tested on a collected external test set of 334 cases. Although the differences in CMR images between high-risk and low-risk MACE patients were not very significant, the method still achieved relatively satisfactory results. Example results are shown below. Figure 2 As shown.

[0032] Quantitative analysis Table 1 shows the prediction results of the DeepSTEMI and Eitel methods of this invention on the external test set.

[0033] Table 1

[0034] The DeepSTEMI model of this invention classified 38 patients (11.4%) as high-risk and 296 patients (88.6%) as low-risk in an external test set. Statistical results are shown in Table 1. The 2-year MACE incidence rates based on DeepSTEMI were 63.2% and 3.7% in the high-risk and low-risk groups, respectively (Chi-square test, P < 0.0001). Compared with the low-risk group, patients in the high-risk group had larger infarct areas (median: 25%, IQR: 18%–34% vs. 21%, IQR: 15%–32%) (P = 0.047), while there were no statistically significant differences in left LVEF, MVO prevalence, and MVO extent between the groups (all P > 0.05).

[0035] A comparison of DeepSTEMI predictive models and Eitel CMR scores. Based on the Eitel CMR risk score, 170 patients (50.9%) in the external testing group were classified as high-risk (Eitel score ≥1), and 164 patients (49.1%) were classified as low-risk (Eitel score <1). The 2-year MACE incidence rates in the high-risk and low-risk groups based on the Eitel score were 12.4% and 8.5%, respectively. There was no difference in the proportion of patients with an Eitel score ≥1 between the high-risk and low-risk DeepSTEMI groups (chi-square test, P=0.816). Table 1 shows the prognostic performance of the DeepSTEMI predictive model and the conventional Eitel CMR score in different subgroups. In the overall external test cohort (n=334), the DeepSTEMI model demonstrated superior predictive ability, with a sensitivity of 0.686 (95% CI: 0.523–0.829), a specificity of 0.973 (95% CI: 0.955–0.990), and an accuracy of 0.943 (95% CI: 0.913–0.967), significantly outperforming the Eitel score (sensitivity 0.600, specificity 0.502, accuracy 0.512). The results indicate that the DeepSTEMI model exhibits robust performance across all assessment categories, including age stratification, sex, CMR time, infarct location, vendor platform, LVEF category, and MVO status, with NRIs ranging from 0.772 to 1.594 (all P < 0.05). Notably, the DeepSTEMI model demonstrated particularly strong accuracy and specificity (accuracy: 0.971, specificity: 0.982) in patients who underwent CMR imaging 4–7 days after infarction, compared to patients who underwent CMR imaging 1–3 days after infarction (accuracy: 0.868, specificity: 0.944).

[0036] Table 2 shows the statistical results of the ablation experiment prediction of the DeepSTEMI method of the present invention on the external test set.

[0037] Table 2

[0038] To illustrate the effectiveness of each module of DeepSTEMI, Table 2 compares the results of ablation experiments for each module. Compared to the results using all three MRI sequences (Cine, LGE, and T2-STIR) and clinical data (age, sex, Grace score, and Killip score), DeepSTEMI using only the imaging sequences showed a sensitivity of 0.651 (95% CI: 0.547–0.802), a specificity of 0.956 (95% CI: 0.931–0.968), and an accuracy of 0.929 (95% CI: 0.897–0.952) on the external test set, which is slightly lower than the results combined with clinical data. Compared to using multi-sequence magnetic resonance imaging, using only movie sequence images for MACE temporal prediction performs poorly, with a sensitivity of 0.535 (95% CI: 0.489–0.620), a specificity of 0.864 (95% CI: 0.822–0.895), and an accuracy of 0.797 (95% CI: 0.735–0.821). It can be seen that the use of multimodal data and multi-sequence magnetic resonance imaging in this invention plays a very important role in improving the accuracy of DeepSTEMI prediction results.

[0039] The above describes the method for predicting two-year MACE events in STEMI patients based on multi-sequence CMR in embodiments of the present invention. The following describes the system for predicting two-year MACE events in STEMI patients based on multi-sequence CMR in embodiments of the present invention: The data acquisition unit is used to acquire multiple CMR sequence images and clinical data of STEMI patients in the acute phase. The preprocessing unit is used to segment and extract the CMR sequence image using a cardiac ROI extraction model to obtain a cardiac region CMR image and preprocess the cardiac region image. The model building and training unit is used to build a predictive model using cardiac region CMR images and clinical data, randomly initialize model parameters, and train the model using training set samples. The prediction unit is used to predict the probability of a patient's MACE event occurring over two years using a trained prediction model.

[0040] The electronic device provided in this embodiment of the invention can vary significantly due to differences in configuration or performance. It may include one or more central processing units (CPUs) (e.g., one or more processors) and memory, and one or more storage media (e.g., one or more mass storage devices) for storing applications or data. The memory and storage media can be temporary or persistent storage. The program stored in the storage media may include one or more modules, each module including a series of instruction operations on the electronic device. Furthermore, the processor may be configured to communicate with the storage media and execute the series of instruction operations stored in the storage media on the electronic device.

[0041] The electronic device may also include one or more power supplies, one or more wired or wireless network interfaces, one or more input / output interfaces, and / or one or more operating systems, such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, etc. Those skilled in the art will understand that the structure of the electronic device does not constitute a limitation on the electronic device itself, and may include more or fewer components than described above, or combine certain components, or have different component arrangements.

[0042] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium, wherein the computer-readable storage medium stores instructions that, when executed on a computer, cause the computer to perform the steps of the aforementioned method.

[0043] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the system, device, or unit described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0044] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0045] 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.

Claims

1. A method for predicting two-year MACE events in STEMI patients based on multi-sequence CMR, characterized in that, Includes the following steps: Acquire multiple CMR sequence images and clinical data of STEMI patients in the acute phase, including age, sex, Grace score, and Killip score; The CMR sequence images were segmented and extracted using a cardiac ROI extraction model to obtain cardiac region CMR images, which were then preprocessed. A predictive model was built using cardiac region CMR images and clinical data. The model parameters were randomly initialized, and the model was trained using training set samples. Use a trained predictive model to predict the probability of MACE events occurring in patients over two years.

2. The method for predicting two-year MACE events in STEMI patients based on multi-sequence CMR according to claim 1, characterized in that, The CMR sequence images include Cine sequence images, LGE sequence images, and T2-STIR sequence images.

3. The method for predicting two-year MACE events in STEMI patients based on multi-sequence CMR according to claim 1, characterized in that, The cardiac ROI extraction model is a U-Net model, which was pre-trained on the public ACDC dataset and fine-tuned on a small-scale multi-CMR sequence dataset containing cardiac region annotations.

4. The method for predicting two-year MACE events in STEMI patients based on multi-sequence CMR according to claim 3, characterized in that, The cardiac ROI extraction model was trained for 50 epochs in both the pre-training and fine-tuning steps, with a batch size of 32.

5. The method for predicting two-year MACE events in STEMI patients based on multi-sequence CMR according to claim 1, characterized in that, In the prediction model, clinical data is embedded through a multilayer perceptron to generate a unified feature vector.

6. The method for predicting two-year MACE events in STEMI patients based on multi-sequence CMR according to claim 1, characterized in that, In the prediction model: Select Video Swin Transformer to extract features from Cine sequence images; Select the Swing Transformer to extract features from LGE and T2-STIR image sequences; The Swin Transformer and Video Swin Transformer employ an alternating mechanism of window-based multi-head self-attention and shifted-window multi-head self-attention. Each CMR sequence is embedded into a unified feature vector of dimension 256, and all extracted features are further summed for two-year MACE event prediction.

7. A two-year MACE event prediction system for STEMI patients based on multi-sequence CMR, characterized in that, The system includes: The data acquisition unit is used to acquire multiple CMR sequence images and clinical data of STEMI patients in the acute phase, including age, sex, Grace score and Killip score; The preprocessing unit is used to segment and extract the CMR sequence image using a cardiac ROI extraction model to obtain a cardiac region CMR image and preprocess the cardiac region image. The model building and training unit is used to build a predictive model using cardiac region CMR images and clinical data, randomly initialize model parameters, and train the model using training set samples. The prediction unit is used to predict the probability of a patient's MACE event occurring over two years using a trained prediction model.

8. An electronic device, the electronic device comprising a memory and at least one processor, the memory storing instructions; The at least one processor invokes the instructions in the memory to cause the electronic device to perform the steps of the two-year MACE event prediction method for STEMI patients based on multi-sequence CMR as described in any one of claims 1-6.

9. A computer-readable storage medium having instructions stored thereon, characterized in that, When the instructions are executed by the processor, they implement the various steps of the two-year MACE event prediction method for STEMI patients based on multi-sequence CMR as described in any one of claims 1-6.

Images