Aortic Cavity Segmentation in PLAX Images Using Deep Learning with Heatmap Guidance

Accurate segmentation of the aortic lumen in the parasternal long-axis (PLAX) view at end-diastole during echocardiography is crucial for providing reliable data for assessing cardiac function and diagnosing cardiovascular diseases. The shape and size of the aortic cavity are closely related to heart function, and accurate segmentation enables clear delineation of the aortic cavity, providing reliable data for heart function evaluation. However, this task presents several challenges due to the dynamic nature of heart motion, variability in aortic anatomy, and interference from surrounding structures, which can complicate segmentation. Moreover, variations in image quality further add to the complexity of obtaining precise segmentations. To address these challenges, we propose a deep learning-based approach that incorporates both the original PLAX image and a heatmap generated using keypoint detection models followed by optical flow techniques. This heatmap, highlighting the aortic valve annulus at the specific end-diastolic phase, provides valuable spatial information that guides segmentation. Our model is built on an enhanced U-Net architecture, with attention mechanisms incorporated within fusion blocks to prioritize relevant regions while suppressing noise from irrelevant areas. This method significantly improves segmentation accuracy by focusing on the critical features of the aortic cavity, while effectively mitigating interference from surrounding structures and motion artifacts. Compared to traditional segmentation methods, our approach outperforms in metrics such as Dice Similarity Coefficient, Mean Intersection over Union, and the 95th percentile Hausdorff Distance, significantly improving the accuracy of aortic cavity segmentation and demonstrating greater potential for practical applications.