Medical image segmentation based on self-supervised hybrid fusion network.

Area of Science:

• Deep learning
• Medical image analysis
• Artificial intelligence in healthcare

Background:

• Accurate medical image segmentation is crucial for disease diagnosis and treatment planning.
• Manual segmentation of brain tumors from MRI is time-consuming and requires expert analysis.
• Computer-aided methods can significantly improve the efficiency and accuracy of medical diagnoses.

Purpose of the Study:

• To develop a self-supervised deep learning network for automated brain tumor segmentation.
• To enhance multi-modal feature extraction and improve model robustness in medical imaging.
• To provide a more efficient and accurate computer-aided diagnosis tool for brain tumors.

Main Methods:

• Designed a multi-modal encoder-decoder network extending the residual network architecture.
• Introduced a multi-modal hybrid fusion module for effective extraction of unique features from each modality.
• Implemented a self-supervised learning approach using a pretext task to complete masked areas, enhancing feature learning and noise immunity.

Main Results:

• The proposed network demonstrated superior performance compared to existing methods on tested datasets.
• The multi-modal hybrid fusion module effectively extracted unique features and reduced framework complexity.
• Self-supervised learning improved the encoder's multi-modal feature extraction capabilities and noise immunity.

Conclusions:

• The developed self-supervised deep learning network offers a promising approach for accurate brain tumor segmentation.
• The method effectively leverages multi-modal MRI data for improved diagnostic accuracy and efficiency.
• This work contributes to advancing computer-aided diagnosis in neuro-oncology.