ACEA-Net: Weakly Supervised Prostate 3D MRI Image Segmentation via Advanced Prompt Points

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Prostate 3D MRI segmentation is crucial for diagnosis and treatment planning.
• Manual slice-wise annotation is time-consuming and requires specialized expertise.
• Existing methods struggle with incomplete or noisy annotations.

Purpose of the Study:

• To develop an efficient and accurate weakly supervised semantic learning framework for prostate 3D MRI segmentation.
• To overcome the limitations of traditional annotation methods and improve segmentation performance.
• To introduce a novel seed expansion strategy and attention mechanisms for enhanced segmentation accuracy.

Main Methods:

• Pseudo-label generation using one foreground seed point and six edge relaxation points.
• Development of ACEA-Net, a weakly supervised semantic learning segmentation framework.
• Implementation of Seed Cluster Geodesic Distance Transform (SeedGeo) for seed expansion.
• Integration of Adaptive Convolutional Normalization (ACN) and Enhanced Simple Parameter-Free Attention Module (SimAM) for label smoothing and noise suppression.
• Utilizing a U-Net baseline model for segmentation.

Main Results:

• Achieved Dice similarity coefficients (Dice) of 87.23% on the MSD prostate dataset and 81.00% on the PROMISE12 prostate dataset.
• Obtained Average Symmetry Surface Distances (ASSD) of 1.73mm and 2.02mm for the respective datasets.
• Demonstrated superior performance compared to current state-of-the-art methods in prostate MRI segmentation.
• Effectively addressed the under-expansion problem in pseudo-labeling through the SeedGeo strategy.

Conclusions:

• ACEA-Net provides an efficient and accurate solution for prostate 3D MRI segmentation.
• The proposed weakly supervised approach significantly reduces the need for extensive manual annotations.
• The integration of SeedGeo, ACN, and SimAM modules enhances segmentation robustness and accuracy.
• This framework represents a significant advancement in automated medical image analysis for prostate cancer assessment.