Attention-Enhanced U-Net for Accurate Segmentation of COVID-19 Infected Lung Regions in CT Scans
A new attention-enhanced U-Net model achieves a Dice score of 0.8658 for identifying infected lung regions in CT scans.
Deep Dive
Researchers Amal Lahchim and Lazar Davic from the University of Kragujevac developed an 'Attention-Enhanced U-Net' model for medical image segmentation. The model, built with PyTorch, uses attention mechanisms and data augmentation to automatically identify COVID-19 infected lung regions in CT scans. It achieved a high Dice coefficient of 0.8658 and a mean Intersection over Union (IoU) of 0.8316, outperforming other methods. This tool can help radiologists quantify lung damage more quickly and consistently.
Why It Matters
This AI could accelerate diagnosis and treatment planning by providing precise, automated analysis of lung CT scans.