Cardiac echocardiographic analysis with multi-scale effective fusion module: a novel stroke prediction approach

Figures

• 

  Fig. 1. Flowchart of patient selection and dataset creation

• 

  Fig. 2. (A) The four-chamber view shows all four heart chambers (left/right atria and ventricles), essential for evaluating chamber size, function, and relationships. (B) The long-axis view highlights the left ventricle, aorta, and valves (mitral and aortic), aiding in the assessment of ventricular function and valve abnormalities. (C) The short-axis view provides cross-sectional images of the heart at various levels, showing structures like papillary muscles and left ventricular wall thickness

• 

  Fig. 3. The risk factors in multivariable logistic regression analysis

• 

  Fig. 4. While the traditional method suffers from weak feature correlations and redundancy, the EFC module incorporates Grouped Feature Focus and Feature Reconstruction to enhance feature representation and recover small object details, resulting in improved detection performance

• 

  Fig. 5. The architecture of the MSEF module. Input feature maps are processed through 1 × 1 convolutions (Conv1 and Conv2) to extract initial representations. Channel Attention selectively enhances important features across channels, while the Grouped Feature Focus refines spatial and grouped information. Simultaneously, Feature Reconstruction recovers lost details and complements the grouped focus. The refined features are further processed by Depth-wise Convolution to extract fine-grained details efficiently, followed by Dynamic Reweighting, which adjusts feature importance dynamically. The final optimized output features are then produced for subsequent network operations

• 

  Fig. 6. Multi-view echocardiographic image classification pipeline. In the Multi-View Feature Extraction stage, features are extracted independently from three echocardiographic views. These features are then fused to integrate spatial and semantic information. The fused features are passed through a ResNet-based Backbone to further refine the representations. Finally, the Classification Head is applied to reduce the feature dimensions and produce the final classification probabilities

• 

  Fig. 7. Model performance on validation set

• 

  Fig. 8. Model performance on test set. The prefix “C” denotes “Clinical Only” models. Conversely, the prefix “M” signifies “Multimodal” models

• 

  Fig. 9. Model attention heatmaps on echocardiographic images for stroke risk prediction. This figure visually demonstrates the regions of highest model attention when assessing stroke risk. Rows 1, 3, and 5 display original echocardiographic images (B-mode and Color Doppler). Rows 2, 4, and 6 show the corresponding images with overlaid Grad-CAM heatmaps, where warm colors (red/orange) indicate high attention and cool colors (blue/green) indicate lower attention. The heatmaps consistently highlight critical cardiac features associated with stroke risk, such as interventricular septal thickening, left ventricular posterior wall thickening, left atrial chamber morphology, and valvular regurgitation