Rapid diagnosis of celiac disease based on serum infrared spectroscopy combined with deep learning

Celiac disease (CD) and its early-stage variant, potential celiac disease (PCD), pose considerable diagnostic challenges due to their overlapping serological and symptomatic features, and the lack of villous atrophy in PCD. Accurate and non-invasive differentiation between CD, PCD, and healthy controls (HC) is crucial for timely intervention and effective management. In this study, we propose a novel hybrid dual-attention deep learning model for the classification of CD, PCD, and HC using Fourier-transform infrared (FTIR) spectroscopy of serum samples. The proposed model integrates two heterogeneous attention mechanisms in a dual-branch architecture: a self-attention module to capture long-range global dependencies across the spectral dimensions, and a channel-wise attention mechanism to highlight locally significant spectral features. The fusion of global and local information enables the model to sensitively detect subtle spectral variations associated with PCD and to robustly discriminate it from both CD and HC, which is a significant advancement over conventional methods. Comprehensive experiments demonstrate that our model achieves superior classification performance compared to baseline architectures, with notably improved sensitivity in distinguishing PCD from CD. This framework offers a promising, non-invasive, and interpretable approach for early and precise diagnosis of celiac spectrum disorders, and may serve as a decision support tool in clinical settings.