MURDA: Multisource\nUnsupervised Raman Spectroscopy\nDomain Adaptation Model with Reconstructed Target Domains for Medical\nDiagnosis Assistance

Artificial intelligence combined with Raman spectroscopy\nfor disease\ndiagnosis is on the rise. However, these methods require a large amount\nof annotated spectral data for modeling to achieve high diagnostic\naccuracy. Annotating labels consumes significant medical resources\nand time. To reduce dependence on labeled medical data resources,\nwe propose a method called Multisource Unsupervised Raman Spectroscopy\nDomain Adaptation Model with Reconstructed Target Domains (MURDA).\nIt transfers knowledge learned from source domain data sets of different\ndiseases to an unlabeled target domain data set. Compared to knowledge\ntransfer from a single source domain, knowledge from multiple disease\nsource domains provides more generalized knowledge. Considering the\ndiversity of autoimmune diseases and the limited sample size, we apply\nMURDA to assist in the medical diagnosis of autoimmune diseases. Additionally,\nwe propose a Double-Branch Multiscale Convolutional Self-Attention\n(DMCS) feature extractor that is more suitable for spectral data feature\nextraction. On three sets of serum Raman spectroscopy data sets for\nautoimmune diseases, the multisource domain adaptation diagnostic\naccuracy of MURDA was superior to traditional single source and multisource\nmodels, with accuracy rates of 73.6%, 83.4%, and 82.9%, respectively.\nCompared with pure source tasks without domain adaptation, it improved\nby 15.1%, 36%, and 21.6%, respectively. This demonstrates the effectiveness\nof Raman spectroscopy combined with MURDA in diagnosing autoimmune\ndiseases. We investigated the important decision dependency peaks\nin migration tasks, providing assistance for future research on artificial\nintelligence combined with Raman spectroscopy for diagnosing autoimmune\ndiseases. Furthermore, to validate the effectiveness and generalization\nperformance of MURDA, we conducted experiments on the publicly available\nRRUFF data set, exploring the application of multisource unsupervised\ndomain adaptation in more Raman spectroscopy scenarios.