Unsupervised Domain Adaptation with Raman Spectroscopy for Rapid Autoimmune Disease Diagnosis

Autoimmune diseases constitute a broadly prevalent category of disorders. Conventional computer-aided diagnostic (CAD) techniques rely on large volumes of data paired with reliable annotations. However, the diverse symptomatology and diagnostic complexity of autoimmune diseases result in a scarcity of reliably labeled biological samples. In this study, we propose a pseudo-label-based conditional domain adversarial network (CDAN-PL) framework by integrating Raman spectroscopy with domain adaptation technology, enabling label-free unsupervised transfer diagnosis of diseases. Compared to traditional unsupervised domain adaptation techniques, our CDAN-PL framework generates reliable pseudo-labels to ensure the robust implementation of conditional adversarial methods. Additionally, its spectral data-adaptive feature extraction techniques further solidify the model's superiority in Raman spectroscopy-based disease diagnosis. CDAN-PL exhibits excellent performance in homologous transfer tasks, achieving an average accuracy of 92.3%-surpassing the baseline models' 80.81% and 86.4%. Moreover, it attains an average accuracy of 90.05% in non-homologous transfer tasks, further validating its generalization capability.