**3.3 Deep learning for classification of OCT images**

OCT is also often used to detect BBC at the dermis-epidermal junction due to its ability to sub-epidermally visualize the skin structure and any underlying lesions [86]. Li et al. [87] used an image-based approach to identify the skin surface and normalized the skin image by surface flattening, then used pre-trained AlexNet, VGG-16, VGG-19 and GoogLeNet for deep feature extraction, and finally used SVM for BCC Classification. The experimental results show that the system based on the VGG-16 image descriptor is the best with a sensitivity of 0.935.

In the current research, image-based deep learning is mainly used in medical image noise reduction and reconstruction processing. Disease diagnosis mainly focuses on tumor detection, brain nervous system disease classification, and cardiovascular disease detection. There are few related studies in the field of skin [88–93]. For example, Kermany D et al. applied deep learning to a dataset of optical coherence tomography images to form a diagnostic tool for screening patients with common treatable blind retinal diseases. The trial results showed that the diagnostic tool was as accurate as hospital specialists in classifying age-related macular degeneration and diabetic macular edema [94]. Abdolmanafi et al. also used intra-coronary tomography images provided by intravascular optical coherence tomography (IV-OCT) as a deep learning library, and extracted the features of convolutional neural network features and fully convolutional network. The accuracy of the diagnostic models can be as high as 90% [95].
