Cross-Modal Prostate Cancer Segmentation via Self-Attention Distillation

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Accurate prostate cancer segmentation from multi-modal magnetic resonance imaging (MRI) is crucial for disease assessment and treatment planning.
• Efficiently utilizing multi-modal MRI features remains a significant challenge in medical image segmentation.

Purpose of the Study:

• To develop an advanced network for automatic and accurate prostate cancer segmentation using multi-modal MRI.
• To enhance the exploitation of intermediate layer information across different MRI modalities.

Main Methods:

• A cross-modal self-attention distillation network was developed.
• The network leverages attention maps for transferring discriminative information between modalities.
• A spatial correlated feature fusion module was incorporated to learn complementary and non-linear information.

Main Results:

• The proposed network effectively utilizes encoded information from intermediate layers of different MRI modalities.
• Attention maps facilitate the transfer of detailed and discriminative information.
• The spatial correlated feature fusion module enhances the learning of complementary image features.
• The model achieved state-of-the-art performance in five-fold cross-validation on 358 biopsy-confirmed MRI images.

Conclusions:

• The developed cross-modal self-attention distillation network offers a robust solution for prostate cancer segmentation.
• The method demonstrates superior performance by effectively integrating multi-modal MRI data.
• This approach holds significant potential for improving prostate cancer assessment and treatment planning.