Abdominal multi-organ segmentation in Multi-sequence MRIs based on visual attention guided network and knowledge...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Abdominal organ segmentation in MRI is crucial for therapy.
• Deep learning methods face challenges with blurred boundaries and low contrast.

Purpose of the Study:

• To propose a novel deep learning network for abdominal multi-organ segmentation using unpaired multi-sequence MRI.
• To address challenges in segmenting organs with indistinct edges and low contrast.

Main Methods:

• Developed a multi-scale visual attention-guided network (VAG-Net).
• Introduced a visual attention-guided (VAG) mechanism to enhance contextual information extraction, especially at organ edges.
• Implemented a knowledge distillation-inspired loss function to reduce semantic differences between MRI sequences.

Main Results:

• VAG-Net achieved superior performance on the CHAOS 2019 Challenge dataset.
• Achieved Dice Similarity Coefficient (DSC) values of 91.83 ± 0.24% (T1-DUAL) and 94.09 ± 0.66% (T2-SPIR).
• Outperformed six state-of-the-art methods in abdominal multi-organ segmentation.

Conclusions:

• The proposed VAG-Net demonstrates superior performance in abdominal multi-organ segmentation.
• The method is particularly effective for segmenting small organs like kidneys.