Raman spectroscopy has proved its capability as an objective, non-invasive tool for the detection of various melanoma and non-melanoma skin cancers (NMSC) in a number of studies. Most publications are based on a Raman microspectroscopic ex vivo approach. In this in vivo clinical evaluation, we apply Raman spectroscopy using a fibre-coupled probe that allows access to a multitude of affected body sites. The probe design is optimized for epithelial sensitivity, whereby a large part of the detected signal originates from within the epidermal layer's depth down to the basal membrane where early stages of skin cancer develop. Data analysis was performed on measurements of 104 subjects scheduled for excision of lesions suspected of being malignant melanoma (MM) (n = 36), basal cell carcinoma (BCC) (n = 39) and squamous cell carcinoma (SCC) (n = 29). NMSC were discriminated from normal skin with a balanced accuracy of 73% (BCC) and 85% (SCC) using partial least squares discriminant analysis (PLS-DA). Discriminating MM and pigmented nevi (PN) resulted in a balanced accuracy of 91%. These results lie within the range of comparable in vivo studies and the accuracies achieved by trained dermatologists using dermoscopy. Discrimination proved to be unsuccessful between cancerous lesions and suspicious lesions that had been histopathologically verified as benign by dermoscopy.

The application of fiber-coupled Raman probes for the discrimination of cancerous and normal skin has the advantage of a non-invasive in vivo application, easy clinical handling, and access to the majority of body sites, which would otherwise be limited by stationary Raman microscopes. Nevertheless, including optical fibers and miniaturizing optical components, as well as measuring in vivo, involves the sensibility to external perturbation factors that could introduce artifacts to the acquired Raman spectra and thereby potentially reduce classification performance. In this study, typical perturbation factors of Raman measurements with a Raman fiber probe, optimized for clinical in vivo discrimination of skin cancer, were investigated experimentally. Measurements were performed under standardized conditions in clinical settings in vivo on human skin, as well as ex vivo on porcine ears. Raman spectra were analyzed in the fingerprint region between 1150 and 1730 cm(-1) using principal component analysis. The largest artifacts in the Raman spectra were found in measurements performed under the influence of strong ambient light conditions as well as after miscellaneous pre-treatments to the skin, such as use of a permanent marker or a sunscreen. Minor influences were also found in measurements using H2O immersion and when varying the probe contact force. The effect of reasonable variation of the fiber-bending radius was found to be of negligible impact. The influence of measurements on hairy or sun-exposed body sites, as well as inter-subject variation, was also investigated. The presented results may serve as a guide to avoid negative effects during the process of data acquisition and so avoid misclassification in tumor discrimination.

No abstract