AI-assisted identification of nonmelanoma skin cancer structures based on combined line-field confocal optical coherence tomography and confocal Raman microspectroscopy.

Abstract

Significance: Morpho-chemical characterization of skin cancers provides valuable insights for early diagnosis, classification, and treatment response assessment.

Aim: We introduce a compact, noninvasive system combining high-resolution morphological imaging and chemical characterization of skin tissues. The system integrates line-field confocal optical coherence tomography for cellular-level imaging and confocal Raman microspectroscopy to analyze the chemical composition of specific targets identified within the morphological images.

Approach: We present results obtained from the system installed in a clinical setting over the course of 1 year. More than 330 nonmelanoma skin cancer specimens were imaged ex vivo, with different structures targeted for Raman microspectroscopy, resulting in over 1300 spectral acquisitions. To evaluate the system's ability to accurately identify cancerous structures, an artificial intelligence model was trained on the spectral data.

Results: The model demonstrated high classification performance, achieving an area under the ROC curve of 0.95 for basal cell carcinoma structures and 0.92 when including structures from both basal and squamous cell carcinomas.

Conclusions: Spectral attention scores derived from Raman data revealed key chemical differences among the various cancerous structures, offering deeper insights into their composition.