Raman Spectral\nFeature Enhancement Framework for Complex\nMulticlassification Tasks

Raman spectroscopy enables label-free clinical diagnosis\nin a single\nstep. However, identifying an individual carrying a specific disease\nfrom people with a multi-disease background is challenging. To address\nthis, we developed a Raman spectral implicit feature augmentation\nwith a Raman Intersection, Union, and Subtraction augmentation strategy\n(RIUS). RIUS expands the data set without requiring additional labeled\ndata by leveraging set operations at the feature level, significantly\nenhancing model performance across various applications. On a challenging\n30-class bacterial classification task, RIUS demonstrated a substantial\nimprovement, increasing the accuracy of ResNet by 2.1% and that of\nSE-ResNet by 1.4%, achieving accuracies of 85.7% and 87.1%, respectively,\non the Bacteria-ID-4 Data set, where RIUS improved ResNet and SE-ResNet\naccuracies by 13.6% and 14.5%, respectively, with only ten samples\nper category. When the sample size was reduced, accuracy gains increased\nto 31.7% and 38.3%, demonstrating the method’s robustness across\ndifferent sample volumes. Compared to basic augmentation, our method\nexhibited superior performance across various sample volumes and demonstrated\nexceptional adaptability to different levels of complexity. RIUS exhibited\nsuperior performance, particularly in complex settings. Moreover,\ncluster analysis validated the effectiveness of the implicit feature\naugmentation module and the consistency between theoretical design\nand experimental results. We further validated our approach using\nclinical serum samples from 70 breast cancer patients and 70 controls,\nachieving an AUC of 0.94 and a sensitivity of 92.9%. Our approach\nenhances the potential for precisely identifying diseases in complex\nsettings and offers plug-and-play enhancement for existing classification\nmodels.