Fuzzy dilated CNN with attention U-Net lung segmentation hits AUC 0.98 for bacterial pneumonia on chest X-ray

A fuzzy dilated CNN with attention-based U-Net lung segmentation achieved an AUC of 0.98 for bacterial pneumonia on chest X-ray. Adding fuzzy logic improved accuracy by 4%, and augmentation boosted AUC by 15% for that class.

• Attention U-Net lung segmentation yielded 2% higher mean Dice compared with U-Net, pruned U-Net, and U-Net++.
• Integration of fuzzy logic into the dilated CNN classifier contributed a 4% increase in accuracy.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.