Andrew Terentis, Ph.D., senior author, professor and chair of Florida Atlantic's Department of Chemistry and Biochemistry.
Photo Credit: Courtesy of Florida Atlantic University
Scientific Frontline: Extended "At a Glance" Summary: Raman Spectroscopy and Artificial Intelligence in Skin Cancer Detection
The Core Concept: A non-invasive diagnostic approach that combines Raman spectroscopy, a technique that captures the molecular fingerprint of tissue, with machine-learning algorithms to detect and classify skin cancer.
Key Distinction/Mechanism: Unlike a conventional biopsy that requires the surgical removal and microscopic examination of tissue, this method utilizes a handheld probe to analyze how light scatters when interacting with cellular molecules. Machine-learning models process the resulting spectral data to identify distinct molecular patterns, successfully distinguishing cancerous lesions, which exhibit stronger protein-related signals, from normal skin, which displays stronger lipid-related signals.
Major Frameworks/Components:
- A mobile Raman spectroscopy system equipped with a 785-nanometer diode laser and a handheld probe.
- Machine-learning classifiers, including K-nearest neighbors, support vector machines, and shallow neural networks, capable of achieving up to 84% test accuracy in classifying tissue.
- Molecular analysis of ex vivo clinical samples, focusing specifically on basal cell carcinoma, squamous cell carcinoma, and normal skin.
Branch of Science: Optical Spectroscopy, Biochemistry, Oncology, Dermatology, and Computer Science.
Future Application: The development of accessible, portable clinical diagnostic tools powered by deep neural networks, enabling rapid, real-time assessment of skin lesions without immediate surgical intervention.
Why It Matters: Skin cancer is the most common cancer worldwide, with 5.4 million nonmelanoma cases diagnosed annually in the United States alone. A highly accurate, non-invasive screening technology could drastically reduce the number of unnecessary, costly, and invasive biopsies while ensuring prompt clinical intervention.
Results of the study, published in the Proceedings of SPIE, showed that the strongest machine-learning models correctly classified the three tissue types approximately 81 to 84 percent of the time. The researchers also found that the models were better at distinguishing cancerous tissue from normal skin than at differentiating basal cell carcinoma from squamous cell carcinoma. The preliminary findings demonstrate promise for a rapid, noninvasive diagnostic tool. Larger studies and more advanced machine-learning techniques are needed to improve accuracy before the approach can be used clinically.
Skin cancer is the most common cancer in the United States and among the most common worldwide. Nearly 1.5 million new cases were diagnosed globally in 2024, including nearly 340,000 melanomas. Nonmelanoma skin cancers—primarily basal cell carcinoma (BCC) and squamous cell carcinoma (SCC)—are even more common, with 5.4 million cases diagnosed annually in the United States.
Early detection is critical. However, distinguishing cancerous lesions from benign and precancerous growths can be challenging because many can appear similar. A biopsy followed by microscopic examination of tissue remains the gold standard for diagnosis, but biopsies are invasive, costly, and can sometimes be performed on lesions that ultimately prove to be benign.
There is a clear need for rapid, noninvasive diagnostic tools that can accurately identify skin cancer while potentially reducing unnecessary biopsies.
Florida Atlantic University researchers are exploring a new way to detect skin cancer by combining Raman spectroscopy—a technique that provides a molecular “fingerprint” of tissue—with machine learning. Unlike a conventional biopsy, Raman spectroscopy analyzes how light scatters when it interacts with molecules in tissue, providing information about its chemical composition without requiring the tissue to be removed or specially prepared.
“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, PhD, senior author, professor, and chair of the Department of Chemistry and Biochemistry at the FAU Charles E. Schmidt College of Science. “By combining the molecular information provided by Raman spectroscopy with machine learning, we are beginning to see how subtle differences in tissue chemistry can be used to distinguish cancer from normal skin.”
To test the approach, researchers used a mobile Raman spectroscopy system equipped with a 785-nanometer diode laser and a handheld probe. They analyzed more than 50 clinical samples ex vivo, including BCC, SCC, and normal skin. They generated nearly 1,000 Raman spectra and assessed a variety of machine-learning methods to determine how accurately the spectral data could distinguish among the three tissue types.
Results of the study, published in the Proceedings of SPIE as part of Advanced Chemical Microscopy for Life Science and Translational Medicine 2026, showed that several machine-learning approaches could identify patterns in Raman spectra that distinguish normal skin from cancerous tissue. K-nearest neighbors and support vector machine classifiers achieved the highest overall test accuracy, at approximately 84 percent. The support vector machine achieved 78.7 percent sensitivity and 88.6 percent specificity, while a shallow neural network achieved 80.8 percent accuracy and the highest receiver operating characteristic area under the curve (ROC AUC) at 0.910.
The researchers first examined the Raman spectra to identify molecular patterns associated with the different tissue types. The analysis showed that normal tissue could be distinguished relatively well from the two types of skin cancer, while BCC and SCC showed greater overlap in their molecular signatures. Raman spectra from the cancerous samples tended to show stronger protein-related signals, whereas normal tissue showed stronger lipid-related signals, providing clues about the molecular characteristics that could help differentiate tissue types.
“Our results are preliminary, but they point toward a future in which a rapid, noninvasive measurement could help guide clinical decisions and potentially reduce unnecessary biopsies,” Terentis said.
The researchers are now looking toward larger studies, further optimization of the machine-learning models, and more advanced approaches, including deep neural networks, which could improve diagnostic performance.
“Raman spectroscopy gives us a wealth of molecular information, but the challenge is teaching a computer to recognize which patterns matter most,” Terentis said. “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.”
The study demonstrates how advances in optical spectroscopy and machine learning could provide a new path toward faster, noninvasive skin cancer assessment, potentially complementing conventional biopsy and helping clinicians make more informed decisions about which lesions require further evaluation.
Published in journal: Proceedings of SPIE
Authors: Andrew C. Terentis, Venkata Dhulipalla, John Strasswimmer, Phuong Nguyen, Lizzie Klein, and Max McCain
Source/Credit: Florida Atlantic University | Gisele Galoustian
Edited by: Scientific Frontline
Reference Number: ongy093026_02
