Artificial Intelligence-Assisted Infrared Spectroscopy and Chemometrics for Enhanced Histopathology Screening of Micro- and Macrocancer Lesions

Accurate detection of micro- and macrocancer lesions remains a critical challenge in histopathology, as conventional hematoxylin and eosin staining requires labor-intensive analysis and is limited in sensitivity toward microscopic foci. Here, we present an artificial intelligence (AI)-assisted workflow integrating Fourier transform infrared (FT-IR) hyperspectral imaging with chemometric modeling for enhanced cancer screening in lung tissues. Using a focal-plane array (128 × 128 pixels with a pixel projection of 5.5 μm × 5.5 μm), hyperspectral maps were generated, enabling biochemical characterization of distinct morphological structures, including bronchial and vascular walls, parenchyma, and neoplastic regions. Histopathological annotations were employed to construct calibration data sets for noncancerous tissues, microcancer lesions, and macrocancer lesions. Discriminant analysis revealed high predictive accuracy across validation strategies, with CORRS-CV (δ = 5) outperforming conventional <i>k</i>-fold and image-based approaches (AUROC = 0.94, accuracy = 97%, and specificity = 98%). This robust performance reflects reduced cross-validation bias and improved generalizability of predictive models. Importantly, FT-IR imaging enabled the detection of both macro- and microlesions consistent with histological references, while also revealing spectral similarities in vascular walls that occasionally led to false-positive predictions. Together, these findings demonstrated that AI-assisted FT-IR chemometrics offers the rapid, label-free, and spatially resolved detection of cancer lesions, complementing standard histopathology by improving sensitivity to microscopic disease and supporting stratification of tumor progression.