An information bottleneck-based optimal transport network for automated diagnosis of spinal diseases

Spinal diseases are common and widely impactful health issues in modern society. With the advancement of computer vision and medical image analysis, image-based automatic recognition and classification of spinal diseases have become research hotspots. However, existing methods often show limited performance in recognizing spinal diseases from complex or low-quality X-ray images. Their performance is easily affected by noise and exposure variations, leading to the extraction of pseudo-features unrelated to the disease. In addition, discrepancies among data sources and imaging conditions result in poor model generalization, making it difficult to adapt to cross-domain variations in clinical applications. To address these challenges, this study proposes an Information Bottleneck-based Optimal Transport Network (IBOTSpine) for automated diagnosis of spinal diseases. The IBOTSpine model introduces an information bottleneck-constrained feature extraction module that effectively captures disease-relevant structural information while suppressing irrelevant noise. Moreover, by incorporating an optimal transport mechanism, the model learns domain-invariant features, thereby reducing the distribution discrepancy between training and testing data and enhancing robustness and generalization across multi-source datasets. Specifically, the model employs a Swin Transformer as the backbone network and jointly optimizes the information bottleneck and optimal transport losses to achieve synergistic improvement in feature extraction, domain adaptation, and classification performance. Experimental results on real spinal X-ray dataset demonstrate that the proposed model outperforms existing methods in classification accuracy, generalization capability, and feature discriminability, validating its effectiveness and application potential in intelligent spinal image diagnosis.