Rapid Diagnosis of Chronic Kidney Disease via Raman Spectroscopy Combined With a Lightweight Mixture of Experts Method

ABSTRACT Chronic kidney disease (CKD) has become a major global public health challenge, affecting over 850 million people, and is expected to become the fifth leading cause of death worldwide by 2040. Among CKD, IgA nephropathy (IgAN) is the most prevalent primary glomerulonephritis globally and a vital cause of end‐stage renal disease (ESRD). Despite kidney biopsy being considered the gold standard for the diagnosis of IgAN, it is expensive to detect and has significant trauma. Thus, it is particularly important to create novel IgAN diagnoses that are speedy, precise, and noninvasive. In this study, we introduced an innovative method of a lightweight mixture of experts, Deep separable convolution–Mixture of Experts–RegNet (Dsc–MoE–RegNet), to diagnose IgAN based on Raman spectra of serum and tears, which utilizes not only the ability of a mixture of experts to optimize performance but also the ability of deep separable convolution to enhance efficiency, achieving a balance between model performance and complexity. Experimental evidence indicates that the Dsc–MoE–RegNet model has 100% accuracy, sensitivity, and specificity for the classification of IgAN patients and healthy controls with an AUC value of 1, outperforming popular machine learning and deep learning methods. This study demonstrated the immense promise of combining serum and tear‐based Raman spectroscopy with the Dsc–MoE–RegNet method for promptly and precisely identifying patients with IgAN.