Detection of local structural distortions in skin due to skin cancer by raman spectroscopy and machine learning

Abstract

A significant portion of the global population is increasingly affected by skin cancer, particularly basal cell carcinoma (BCC), which is common and often appears on sun-exposed skin areas. BCC typically grows slowly and invades locally but can metastasize. In the U.S., BCC has an incidence rate of 300 per 100,000 people, with a 30% lifetime risk and an annual increase of over 10%. BCC significantly impacts healthcare systems. Modern biomedical research, using technologies like Raman spectroscopy combined with machine learning (ML), offers new ways to accurately diagnose skin malignancies. This study used Raman spectroscopy and supervised ML algorithms to identify structural irregularities in tumor-affected tissues and differentiate between nodular and infiltrative BCC, aiming to speed up diagnosis and reduce mortality. Approved by the ethics committee of Şisli Hamidiye Etfal Training and Research Hospital in Istanbul, Türkiye, the study involved excising BCC tissues for Raman spectroscopy analysis. The data underwent preprocessing steps such as cosmic ray elimination, fluorescence background removal, and spectrum normalization. Nine different ML models were used to classify BCC and differentiate its subtypes. Results showed that RF and KNN achieved the highest accuracy in distinguishing BCC from normal tissue (98.4%) and in differentiating BCC subtypes. This study highlights the potential of Raman spectroscopy and ML as effective, non-invasive tools for diagnosing BCC and its subtypes.