Multimodal Deep Learning Framework Detects Silicosis on Chest X-Ray with 98.73% Accuracy

A fused deep learning model (EfficientNet-B3 + capsule network + ConvNext V2) classified silicosis on chest radiographs with 98.73% accuracy on a single dataset. Not yet externally validated; real-world performance unclear.

• Retrospective study using the Silicodata dataset only; no external or prospective validation cohort reported, limiting generalizability.
• Framework combines three convolutional architectures via feature fusion plus a bidirectional attention classifier; outperformed individual baseline models, but comparators and dataset size are not specified in the abstract.
• Designed as a computer-aided screening tool for resource-limited occupational health settings where expert radiologist access is scarce — a plausible use case, but clinical deployment would require prospective validation.

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