PCNet: Prior Category Network for CT Universal Segmentation Model.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Accurate segmentation of anatomical structures in Computed Tomography (CT) images is essential for clinical applications.
• Current deep learning segmentation methods face limitations due to data scale and model size.
• Clinical diagnosis, treatment planning, and disease monitoring rely heavily on precise CT image segmentation.

Purpose of the Study:

• To develop a novel deep learning approach, the Prior Category Network (PCNet), to improve CT image segmentation.
• To leverage prior knowledge between anatomical categories to boost segmentation performance.
• To create a high-performance, universal model for CT segmentation with enhanced accuracy and robustness.

Main Methods:

• Proposed PCNet framework with three components: prior category prompt (PCP), hierarchy category system (HCS), and hierarchy category loss (HCL).
• PCP utilizes Contrastive Language-Image Pretraining (CLIP) and attention modules to define relationships between anatomical categories.
• HCS guides segmentation through hierarchical relationships, while HCL enforces consistency and directional guidance.

Main Results:

• PCNet demonstrated high-performance and universal applicability for CT segmentation across extensive experiments.
• The framework showed significant transferability on multiple downstream tasks.
• Ablation experiments confirmed the critical importance of the HCS methodology.

Conclusions:

• PCNet effectively enhances CT image segmentation by incorporating prior anatomical knowledge.
• The proposed method offers a robust and transferable solution for medical image analysis.
• The PCNet framework represents a significant advancement in deep learning for medical imaging segmentation.