Automated Classification of Rheumatoid Arthritis and Knee Synovitis From Hyperspectral Reflectance Data

Accurate differentiation between rheumatoid arthritis (RA) and knee synovitis (KS) is essential for guiding optimal treatment, yet conventional histopathology often relies on subjective interpretation and offers limited insight into tissue biochemistry. Here, we introduce TransCNN, a novel multimodal framework that integrates hyperspectral imaging (HSI) with deep learning to achieve objective, high-precision diagnosis. Reflectance-mode HSI across the 400-1000 nm spectrum was performed on 95 synovial tissue specimens. Spectral data were denoised using Savitzky-Golay filtering and distilled via principal component analysis to enhance feature separability. TransCNN employs convolutional neural networks to capture intricate spatial morphology and Transformer layers to model global spectral correlations, producing a unified spectral-spatial representation. On an independent validation set, TransCNN achieved 91% accuracy, 89% F1-score, 90% recall, and 89% precision, substantially surpassing traditional approaches. These findings demonstrate that TransCNN provides a noninvasive, highly sensitive tool for pathological diagnosis, facilitating more reliable, data-driven decision-making in rheumatologic practice.