Keratinocyte Carcinoma Detection via Convolutional Neural Networks

Abstract

Skin cancer is the most prevalent form of cancer. Melanoma and non-melanoma, also known as keratinocyte carcinoma, skin cancers have frequent occurrence although melanoma skin cancer is known to be more deadly. Still, keratinocyte carcinoma skin cancers are encountered with higher frequency and come with more numerous types than melanoma. In this paper, an automated method is used to detect the frequently occurring keratinocyte carcinoma skin cancer. The method is based on deep learning, where AlexNet, ResNet-18, and ResNet-50 architectures are employed to classify common malignant pigmented skin lesion images as belonging to basal cell carcinoma, squamous cell carcinoma or keratinocyte carcinoma. A public archive of skin images is used to test and validate the success of the deep learning methods employed. The results show that ResNet-50 architecture gives the best detection results where for keratinocyte carcinoma detection the area under the receiver operating characteristic curve performance of it is 0.80.