You Need Glimpse Before Segmentation: Stochastic Detector-Actor-Critic for Medical Image Segmentation.

Area of Science:

• Medical Imaging
• Computer Vision
• Artificial Intelligence

Background:

• Medical images often have extensive background noise, hindering segmentation accuracy.
• Current methods may struggle with efficiency and parameter count in complex medical image segmentation tasks.

Purpose of the Study:

• To develop a novel framework, Stochastic Detector-Actor-Critic (SDAC), for efficient and accurate medical image segmentation.
• To address challenges posed by background redundancy and improve performance in low-resource settings.

Main Methods:

• Implemented a Stochastic Detector-Actor-Critic (SDAC) framework integrating a detector network and Actor-Critic algorithm.
• Employed policy gradient algorithms for initial background filtering and pixel-wise mask generation.
• Jointly trained both detector and segmentation modules to minimize error propagation.

Main Results:

• SDAC achieved competitive segmentation performance (DICE and IoU metrics) compared to state-of-the-art methods.
• The framework utilizes 10x fewer parameters than the best-performing baseline.
• Demonstrated robust performance in low-resource settings (50-shot and 100-shot).

Conclusions:

• SDAC offers a lightweight and efficient solution for medical image segmentation.
• The proposed method is suitable for real-world applications, especially in data-scarce environments.
• SDAC serves as an excellent baseline for future medical image segmentation research.