Fuzzy deep learning framework boosts chest X-ray classification accuracy across magnifications

A fuzzy-neural framework with entropy-guided feature selection classified multi-disease chest X-rays with accuracy up to 0.9859 and F1-score up to 0.9889 at ×20 magnification, outperforming standard CNNs by 3.5–8.6% in accuracy.

• The model integrated a ResNet-50 backbone with spatial-channel attention, radiology-aware augmentation, and entropy-guided recursive feature elimination achieving >85% dimensionality reduction.
• A fuzzy-neural classifier with confidence-weighted defuzzification explicitly estimated uncertainty and stabilized predictions in borderline cases.
• Performance was evaluated across four public datasets and four magnification levels (×1 to ×20), maintaining high accuracy even at the lowest resolution (×1: accuracy 0.9401; F1-score 0.9564).

Automated summary

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