In vivo nonmelanoma skin cancer diagnosis using Raman microspectroscopy

Abstract

Background and objectives: Nonmelanoma skin cancers, including basal cell carcinoma (BCC) and squamous cell carcinoma (SCC), are the most common skin cancers, presenting nearly as many cases as all other cancers combined. The current gold-standard for clinical diagnosis of these lesions is histopathologic examination, an invasive, time-consuming procedure. There is thus considerable interest in developing a real-time, automated, noninvasive tool for nonmelanoma skin cancer diagnosis. In this study, we explored the capability of Raman microspectroscopy to provide differential diagnosis of BCC, SCC, inflamed scar tissue, and normal tissue in vivo.

Study design: Based on the results of previous in vitro studies, we developed a portable confocal Raman system with a handheld probe for clinical study. Using this portable system, we measured Raman spectra of 21 suspected nonmelanoma skin cancers in 19 patients with matched normal skin spectra. These spectra were input into nonlinear diagnostic algorithms to predict pathological designation.

Results: All of the BCC (9/9), SCC (4/4), and inflamed scar tissues (8/8) were correctly predicted by the diagnostic algorithm, and 19 out of 21 normal tissues were correctly classified. This translates into a 100% (21/21) sensitivity and 91% (19/21) specificity for abnormality, with a 95% (40/42) overall classification accuracy.

Conclusions: These findings reveal Raman microspectroscopy to be a viable tool for real-time diagnosis and guidance of nonmelanoma skin cancer resection.

Figure 1

Schematic of Raman microspectrometer used for in vivo skin measurements. Handheld probe is fiber coupled to laser and spectrometer. BP: bandpass filter, DM: dichroic mirror, LP: longpass filter, CM: concave mirror.

Figure 2

Mean Raman spectra of skin pathologies studied, normalized to mean intensity for direct comparison.

Figure 3

Statistical differences between the pathologic and normal spectra determined by standard error confidence intervals. Gray bands indicate the 99% confidence intervals of pathologic difference spectra (spectraset minus respective normal).

Figure 4

Posterior probability distributions for all samples studied. Perilesional normal spectra are shown in top figure, and pathologic spectra are shown in lower; labels above each plot correspond to histopathology, while the shading reveals the posterior probability for each pathological classification. Arrows indicate the two misclassified spectra.