Photonic Data Science

Being among the most common cancers worldwide screening and early diagnosis of colorectal cancer is of high interest for the health system, the patients and for research. Raman microspectroscopy as a label-free, noninvasive and non-destructive technique is a promising tool for an early diagnosis. However, to ensure a reliable diagnosis specially designed statistical analysis workflows are required. Several statistical approaches have been introduced leading to varying results in the overall accuracy, sensitivity and specificity. In this study a systematic evaluation of different statistical analysis approaches has been performed using a colon cancer mouse model with genotypic identical individuals. Based on the inter-individual Raman spectral variances a measure for the biological variance can be estimated. By applying a leave-one-individual-out cross-validation a clinically relevant discrimination of healthy tissue versus adenoma and carcinoma with an accuracy of 95% is shown. Furthermore, the transfer of a model from tissue to biopsy specimen is demonstrated.