Point-Supervised Coronary Semantic Segmentation in X-Ray Angiographic Images.

Area of Science:

• Medical Imaging
• Computer-Aided Diagnosis
• Artificial Intelligence in Healthcare

Background:

• Coronary semantic segmentation in X-ray angiography is crucial for diagnosing and planning treatments for coronary artery disease (CAD).
• Manual pixel-level annotation for this task is labor-intensive and challenging due to complex vascular structures and similar branch appearances.
• Existing methods struggle with sparse point-based supervision, often leading to overfitting and poor generalization.

Purpose of the Study:

• To develop a point-supervised coronary semantic segmentation framework that minimizes annotation burden while maintaining high accuracy.
• To address the challenges of overfitting and limited generalization associated with sparse point labels.
• To improve the perception of coronary topology and differentiation between vascular branches.

Main Methods:

• Proposed a point-supervised framework for coronary semantic segmentation.
• Introduced an adaptive foreground mask generation module and region regularization to enrich supervision signals from sparse point labels.
• Developed a multi-task learning framework combining keypoint detection and semantic segmentation using a shared encoder and task-specific decoders.

Main Results:

• The point-supervised model achieved segmentation accuracy comparable to fully supervised methods.
• The proposed framework demonstrated superior performance compared to existing state-of-the-art point-supervised semantic segmentation techniques.
• Effectively reduced the need for dense, pixel-level manual annotations.

Conclusions:

• The novel point-supervised approach significantly reduces annotation effort for coronary semantic segmentation.
• The method offers a viable alternative to fully supervised techniques, achieving comparable performance.
• This framework enhances the feasibility of computer-aided diagnosis for coronary artery disease.