TSRL-Net: Target-aware supervision residual learning for stroke segmentation.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neurology

Background:

• Accurate stroke segmentation in MRI is vital for computer-aided diagnosis of brain diseases.
• Challenges include class imbalance and intraclass ambiguities, leading to false negatives and inaccurate segmentation.

Purpose of the Study:

• To propose a novel target-aware supervision residual learning framework for improved stroke segmentation.
• To address challenges of class imbalance and intraclass ambiguities in MRI stroke segmentation.

Main Methods:

• Developed a target-aware loss function to focus on positive samples and compensate for negative sample losses.
• Implemented a coarse-grained residual learning module to correct lost residual features during decoding.
• Utilized a reverse/positive attention unit to suppress noise and highlight important features.

Main Results:

• The proposed framework demonstrated effectiveness in stroke segmentation on public datasets.
• Achieved superior performance compared to several state-of-the-art methods in experimental evaluations.

Conclusions:

• The novel framework successfully mitigates issues of class imbalance and intraclass ambiguities in stroke segmentation.
• The method offers a promising approach for enhancing computer-aided diagnosis systems for brain diseases.