AI and the World's Hard Problems · Medicine & drug discovery · Nature (computational biology) · 2026-10-18
Multi-task deep learning framework for brain tumor classification and segmentation using ensemble feature extraction
Early and accurate diagnosis of brain tumors is essential for treatment planning, but manual MRI review is slow and prone to human error, and separate classification and segmentation add computational cost. The authors propose a multi-task deep learning framework that combines ResNet-50, EfficientNet-B5, and CNN features with attention modules, fused through dynamic weighting. It classifies first and runs U-Net++ segmentation only when a tumor is detected, and on a public MRI dataset it reported 99.84% classification accuracy, a 91.22% Dice coefficient, and an 89.01% IoU.