Elastic scattering spectrum fused with Raman spectrum for rapid classification of colorectal cancer tissues.

Currently, HE staining and microscopic imaging are the main approaches for the diagnosis of cancerous tissues, which are inefficient, and the results are heavily dependent on doctors' experience. Therefore, establishing a rapid and accurate method for identifying cancerous tissues is of great value for the preoperative and intraoperative assessments. Raman spectroscopy is a non-destructive, label-free and highly specific method, and it has been widely reported in cancer tissue research. However, the low accuracy of Raman spectral results due to the complex compositions of the tissues limits the clinical applications of Raman spectroscopy. In this study, two-dimensional features of the biochemical composition and morphological structure were combined to classify colorectal cancer tissue by innovatively fusing the elastic scattering spectrum and Raman spectrum. In this study, the elastic scattering spectrum and Raman spectrum of 20 clinical colorectal tissues were acquired using a Raman spectrometer and a homemade elastic scattering light device. After multi-modal spectrum data processing and fusion, a composite AI model called spec-transformer was trained and tested. The results showed that the new model classified colorectal tissues with an accuracy of ≥97%. Moreover, Grad-CAM technology was applied to analyse the compositional variation between normal and colorectal cancer tissues, and it demonstrated a high expression of tryptophan and unsaturated fatty acids in cancer tissues with a reduction in tyrosine and beta-carotene expression. Our approach has potential for colorectal cancer diagnosis and could be extended for diagnosis and research on other cancers.