Authors: Ziya Ata Yazıcı, İlkay Öksüz, Hazım Kemal Ekenel

Published on: March 15, 2024

Impact Score: 7.6

Arxiv code: Arxiv:2403.09942

Summary

• What is new: A new multi-scale, attention-guided hybrid U-Net-shaped model called GLIMS is introduced for 3D brain tumor segmentation.
• Why this is important: Glioblastoma is hard to diagnose early due to its varied appearance, and existing automated detection methods struggle with this heterogeneity.
• What the research proposes: GLIMS uses multi-scale feature extraction, Swin Transformer blocks for improved global feature extraction, and attention-guided segmentation mask generation with hierarchical supervision.
• Results: The model achieved 92.19, 87.75, and 83.18 Dice Scores for Whole Tumor, Tumor Core, and Enhancing Tumor regions, respectively, on the validation set.

Technical Details

Technological frameworks used: Multi-scale, attention-guided hybrid U-Net-shaped architecture

Models used: Swin Transformer

Data used: Clinically acquired MRI data from the BraTS challenge

Potential Impact

Healthcare and medical imaging companies, particularly those involved in brain tumor diagnosis and treatment, might see significant benefits or disruption from the adoption of GLIMS.

Want to implement this idea in a business?

We have generated a startup concept here: NeuroAID.