Attention-gated U-Net model for semantic segmentation of brain tumors and feature extraction for survival prognosis
A new AI model achieves 90% accuracy in segmenting brain tumors and predicts patient survival from MRI scans.
Deep Dive
Researchers led by Mohendra Roy developed an Attention-Gated Recurrent Residual U-Net (R2U-Net) model for brain tumor analysis. The model achieved a Dice Similarity Score of 0.900 for Whole Tumor segmentation on the BraTS2021 dataset. It also extracts 64 features to predict patient survival days, achieving 45.71% accuracy. This tool can help radiologists by automating complex tumor segmentation and providing prognostic data to aid in treatment planning.
Why It Matters
Automates a critical, time-consuming medical task, potentially speeding up diagnosis and personalizing treatment plans for glioma patients.