A Multi-Scale Residual Network Based on Kolmogorov–Arnold Networks Combined With Raman Spectroscopy for Rapid Diagnosis of Membranous Glomerulonephritis

Abstract

Membranous nephropathy (MN) is a common cause of nephrotic syndrome in adults and a frequent contributor to end-stage renal disease (ESRD). In recent years, the prevalence of MN has shown an upward trend, with a notably increasing incidence in younger populations. Currently, methods such as renal biopsy and renal function tests are used to diagnose MN. Conventional diagnostic methods carry risks of infection and other complications, in addition to being costly and requiring advanced technical expertise. As a result, the early detection of MN necessitates the development of a diagnostic method that is quick, inexpensive, and noninvasive. This paper proposes a multi-scale residual network (MSRKan) based on Kolmogorov–Arnold networks (KANs) for processing Raman spectroscopic data obtained from the serum of MN patients. The model captures both large-scale global information and fine-grained local details of spectral data, minimizing information loss and enhancing performance. Compared with traditional models, MSRKan achieves the highest accuracy (98.18%), with precision, recall, and F1-score of 100%, 96.67%, and 98.31%, respectively. Additionally, this study verifies for the first time the effectiveness of KAN in spectral data processing. These results demonstrate that the combination of the MSRKan algorithm and Raman spectroscopy enables rapid diagnosis of MN, which holds significant clinical value for patients and enhances the accuracy of computer-aided medical diagnosis. The source code for the MSRKan model is publicly available on GitHub at: https://github.com/cj764/msrkan.