A deep neural network using modified EfficientNet for skin cancer detection in dermoscopic images

Artificial intelligence (AI) systems can assist in analyzing medical images and aiding in the early detection of diseases. AI can also ensure the quality of services by avoiding misdiagnosis caused by human errors. This study proposes a deep neural network (DNN) model with fine-tuned training and improved learning performance on dermoscopic images for skin cancer detection. A knowledge base for training the DL models is constructed by combining different dermoscopic datasets. Transfer learning and fine-tuning are implemented for faster training of the proposed model on a limited training dataset. The data augmentation techniques are applied to enhance the performance of the model. A total of 58,032 refined dermoscopic images were used in this study. The output of the layered architecture is aggregated to perform the binary classification for skin cancer. The performance of the trained models is investigated for multiclass and binary classification tasks. The performance metrics confirm that the DNN network with modified EfficientNetV2-M outperforms the state-of-the-art deep learning-based multiclass classification models.