Separating Skin Cancer From Normal Tissue Using Handheld Raman Spectroscopy

Handheld instruments are becoming more widely used in analytical spectroscopy, and this includes in clinical analysis. A recent study conducted by a team of researchers at Florida Atlantic University (FAU) added to the growing literature of applying handheld Raman spectroscopy in clinical diagnostics.¹ In their study, the team of researchers, led by Andrew Terentis, Ph.D., senior author and professor and chair of the Department of Chemistry and Biochemistry in FAU's Charles E. Schmidt College of Science, paired a handheld Raman spectroscopy system with machine-learning (ML) classifiers to distinguish between normal skin and two common nonmelanoma skin cancers with increased test accuracy of approximately 84%.^1,2

This study represents an early step toward a non-invasive tool that could reduce the number of biopsies. The researchers analyzed nearly 1,000 Raman spectra in the study, collecting these samples ex vivo from more than 50 clinical samples of normal skin, basal cell carcinoma (BCC), and squamous cell carcinoma (SCC).²

Why is studying skin cancer important?

Studying skin cancer is important because of how many individuals and families it impacts annually. Skin cancer is the most common type of cancer in the United States, and nonmelanoma skin cancers, mainly BCC and SCC, account for 5.4 million cases diagnosed in the U.S. each year.² When it comes to treatment, patients normally undergo biopsies followed by microscopic examination of the tissue.² While this treatment method is fairly effective, it is invasive and expensive.² As a result, researchers are exploring new methods that can conduct optical measurements quicker and expedite treatment.

Raman spectroscopy, because it measures how light scatters off molecules in tissue, can reveal the tissue's chemical composition without removing or preparing a sample. This makes it a promising technology in this space.²

"The promise of this technology is that it could give clinicians another way to look beneath the surface of a skin lesion without immediately having to remove tissue," said Andrew Terentis, Ph.D., senior author and professor and chair of the Department of Chemistry and Biochemistry in FAU's Charles E. Schmidt College of Science.²

What were the results of this study?

The team collected spectra using a mobile Raman system equipped with a 785-nm diode laser and a handheld probe.¹ They then compared several traditional ML classifiers on the resulting data set.

Related Content: https://www.spectroscopyonline.com/view/can-wearable-sensor-platforms-detect-cancer-drug-contamination-on-gloves-

Partial least squares discriminant analysis (PLS-DA) produced relatively few false positives and reached a specificity of 95.5%, but its sensitivity was only 66.5%.¹ Principal component analysis combined with quadratic discriminant analysis (PCA-QDA) produced more false positives, mostly from confusing normal tissue with SCC. It reached a specificity of 77.7% and a sensitivity of 77.8%.¹ Both methods had considerable difficulty separating BCC from SCC.

What were the models that performed the best?

The two best-performing models were k-nearest neighbors (KNN) and support vector machine (SVM) classifiers. Each reached the highest overall test accuracy, about 84%, on the three-class task.^1,2 A wide, shallow feedforward neural network achieved 80.8% test accuracy and the highest receiver operating characteristic area under the curve (ROC AUC) of any model tested, at 0.910.^1,2 However, the researchers acknowledged in their study that robust clinical performance would require accuracy, sensitivity, and specificity well above 90%.^1,2

KNN, SVM, and the neural network were considerably better at distinguishing BCC from SCC spectra, and SVM was the strongest of the three on that task.^1,2 Across all three classes, KNN reached 80.0% sensitivity and 87.3% specificity, and SVM reached 78.7% sensitivity and 88.6% specificity.^1,2

"Our results are preliminary, but they point toward a future in which a rapid, non-invasive measurement could help guide clinical decisions and potentially reduce unnecessary biopsies," Terentis said.²

What are the next steps in this research?

The researchers plan larger studies and further tuning of the current models. They also intend to evaluate more advanced approaches, including deep neural networks trained on larger numbers of samples.²

"Raman spectroscopy gives us a wealth of molecular information, but the challenge is teaching a computer to recognize which patterns matter most," Terentis said.² "With more samples and better-trained models, we believe there is significant potential to improve the accuracy of this approach. Ultimately, we want to develop technology that is not only accurate, but also practical, portable and accessible enough to become a useful tool in the clinical setting."

References

1. Terentis, A.; Dhulipalla, V.; Strasswimmer, J.; Nguyen, P.; Klein, L.; McCain, M. Raman Spectroscopy-based Optical Diagnostics of Skin Cancer: Evaluation of ML Classifiers for SCC, BCC, and Normal Tissue. Conf. Proceed. 2026, 13864. DOI: 10.1117/12.3082236
2. Galoustian, G. FAU Study Uses Raman Spectroscopy and AI to Detect Skin Cancer. Florida Atlantic University, 2026. https://www.fau.edu/newsdesk/articles/skin-cancer-raman-spectroscopy-ai (accessed October 1, 2026).