Unimodal and bimodal classification methods for breast carcinomas based on laser-induced autofluorescence...

Despite its efficiency, real-time capability, and low cost, laser-induced fluorescence spectroscopy has limited classification accuracy in breast carcinoma diagnosis, restricting its clinical application. To address this, we evaluated three steady-state autofluorescence analysis approaches: spectral ratio, piecewise linear fitting, and univariate cubic polynomial fitting. To make the analysis more systematic and objective, a sliding-window mechanism and statistical difference analysis were introduced. Based on these results, a novel composite-feature strategy was proposed through arithmetic combination of slope values across multiple spectral segments, enabling distributed spectral information to be effectively integrated and enhancing classification stability. Under the current sample size, piecewise linear fitting demonstrated superior performance by extracting multiple slope features, one of which achieved 100 % classification. However, ambiguous boundaries limited unimodal generalizability for larger datasets. We then combined the optimal unimodal method-piecewise linear fitting-with time-resolved fluorescence lifetime data for bimodal analysis. This fusion markedly enhanced class separability over unimodal approaches. Furthermore, decision boundaries from support vector machines (SVM) were sharper than those from linear discriminant analysis (LDA). These findings highlight the diagnostic value of spectral slope features and emphasize the enhanced classification performance achieved by combining steady-state with time-resolved data. The proposed method is label-free, low-cost, time-efficient, and shows strong potential for intelligent diagnostics by reducing reliance on subjective interpretation. The observed differences likely arise from structural protein changes (collagen, elastin), altered protein-bound NAD(P)H ratios, elevated FAD, and porphyrin accumulation, reflecting tumor-related metabolic and microstructural changes.