Deep Learning Model with Atrous Convolutions for Improving Skin Cancer Classification

Skin cancer is the most common problem all over the world, and some forms of skin cancer are not as aggressive as melanoma. It is vital to identify the type of skin cancer whether benign or malignant for providing timely treatment to the patients to increase the survival rate. The proposed work aims to address this task by proposing a convolutional neural network (CNN) model by leveraging the architecture of a network named EfficientNetB0. The network is fine-tuned with the optimized selection of hyperparameters and network layers are modified to make it suitable for the given dataset. Moreover, the atrous dilated convolution rate is added to some of the feature extraction layers of the existing network. The outcome of the network is analyzed using the locally interpretable model-agnostic explanation (LIME) technique to verify whether the proposed network learned suitable features from the lesion region in different skin cancer images. The proposed model employed three datasets of skin cancer; International Skin Imaging Challenge (ISIC), PH \(^{2}\) , and MED-NODE. It is concluded from the experimental results that the adopted deep neural model with the proposed modification is effective in classifying forms of cancer.