Skin lesion image classification method based on extension theory and deep learning

A skin lesion is a part of the skin that has abnormal growth on body parts. Early detection of the lesion is necessary, especially malignant melanoma, which is the deadliest form of skin cancer. It can be more readily treated successfully if detected and classified accurately in its early stages. At present, most of the existing skin lesion image classification methods only use deep learning. However, medical domain features are not well integrated into deep learning methods. In this paper, for skin diseases in Asians, a two-phase classification method for skin lesion images is proposed to solve the above problems. First, a classification framework integrated with medical domain knowledge, deep learning, and a refined strategy is proposed. Then, a skin-dependent feature is introduced to efficiently distinguish malignant melanoma. An extension theory-based method is presented to detect the existence of this feature. Finally, a classification method based on deep learning (YoDyCK: YOLOv3 optimized by Dynamic Convolution Kernel) is proposed to classify them into three classes: pigmented nevi, nail matrix nevi and malignant melanomas. We conducted a variety of experiments to evaluate the performance of the proposed method in skin lesion images. Compared with three state-of-the-art methods, our method significantly improves the classification accuracy of skin diseases.

Automated summary

In this paper, a two-phase classification method for skin lesion images in Asians is proposed, which integrates medical domain knowledge, deep learning, and a refined strategy. A skin-dependent feature is introduced to efficiently distinguish malignant melanoma, and a classification method based on deep learning is proposed. The proposed method significantly improves the classification accuracy of skin diseases compared to state-of-the-art methods.