Selective Deeply Supervised Multi-Scale Attention Network for Brain Tumor Segmentation.

Area of Science:

• Medical imaging analysis
• Artificial intelligence in oncology
• Neurosurgery and oncology

Background:

• Brain tumors are aggressive cancers requiring timely diagnosis.
• Manual segmentation of brain tumors is labor-intensive and error-prone.
• Automated segmentation methods face challenges due to tumor heterogeneity.

Purpose of the Study:

• To develop a fully automated brain tumor segmentation method.
• To improve the accuracy of segmenting whole, core, and enhancing tumor regions.
• To address the limitations of manual segmentation in clinical workflows.

Main Methods:

• Proposed a selective deeply supervised multi-scale attention network (SDS-MSA-Net).
• Utilized 3D and 2D inputs for sequential information and feature extraction.
• Incorporated multi-scale architecture with attention units and selective deep supervision.

Main Results:

• Achieved improved performance in brain tumor region segmentation on the BraTS2020 dataset.
• Demonstrated particular effectiveness in segmenting core and enhancing tumor subregions.
• The proposed SDS-MSA-Net shows significant potential for clinical application.

Conclusions:

• The SDS-MSA-Net offers an effective and automated solution for brain tumor segmentation.
• The method's ability to handle tumor heterogeneity enhances diagnostic accuracy.
• Public availability of the code facilitates further research and development.