Abstract
Background: Vibrational optical coherence tomography (VOCT) is a new digital technique that identifies cancerous cells, small friable blood vessels, and fibrotic tissues that are the hallmarks of epithelial cell derived skin cancers. This technique uses infrared light and acoustic sound to characterize the components of skin and skin lesions and the differences that occur during cancer. Initial studies focused on understanding the physiological changes that occur that are the basis for noninvasive classification of skin cancers.
Objective: To collect additional clinical optical coherence tomography (OCT) images of cancerous lesions to develop convolutional neural network models of basal cell carcinomas (BCCs), squamous cell carcinomas (SCCs), and melanomas with high sensitivities and specificities. The goal of this research is to produce digital images and AI models that can be used with telemedicine for remote classification of skin cancers.
Methods: OCT images of cancerous lesions were collected using the OptoScope and compared to images of normal skin using a convolutional neural network. The models are developed by comparing the intensity distribution and location of pixels contained in the OCT images.
Results: Gray scale and color-coded OCT images of normal skin, BCCs, SCCs and melanomas were evaluated using convolutional neural network models. The models predicted the type of lesions with sensitivity and specificity greater than 90%. These models were able to predict the correct clasifications of BCC, SCC, and melanomas consistent with the diagnoses based on lesion histopathology conducted by board certified dermatopathologists.
Conclusions: Results of OCT studies indicate that CNN models can classify different skin cancers in several minutes based on OCT images of BCC, SCC, and melanomas. Noninvasive use of the OptoScope provides digital information about skin cancers that can be used to classify skin cancers as part of an evaluation of patient lesions via telemedicine. This is particularly useful in areas where Dermatologists are in short supply and patient waiting times are prolonged. Evaluation using new digital techniques and the availability of approved topical treatments for these cancers will provide an effective manner to treat patients in rural areas where Dermatology visits are not readily available.