DePicT Melanoma Deep-CLASS: a deep convolutional neural networks approach to classify skin lesion images

摘要

Although case-based reasoning (CBR) has been applied in a number of medical systems, only a few systems have been developed for melanoma. The estimated 5-year survival rate for patients whose melanoma is detected early is about 99% [ ] in the USA and around 93% [ ] in Germany. The survival rate falls to 63% when the disease reaches the lymph nodes and 20% when the disease metastasizes to distant organs [ ]; therefore, having the right information at the right time by developing decision support systems is essential and has become a major area of research in this field. The best path to early detection is recognizing new or changing skin growths, especially those that appear different from other moles [ ]. Even after treatment, it is very important for patients to keep up on their medical history and records. The national comprehensive cancer network (NCCN) which is an alliance of 28 cancer centers in the United States, creates helpful reports and resources to serve as guidelines for informing patients and other stakeholders about cancer [ ]. In this paper, we propose a hybrid CBR system and evaluate its performance on the skin lesions classification (benign and malignant). In the proposed system, deep neural networks are used as an image classifier in the context of CBR methodology and in the retrieval process. The case base of our proposed system contains melanoma images and skin cancer information as a case description and recommendation and the structure of cases includes image features, identified keywords, and a word association profile, which is explained in the next section. The case-base updates and learns over the time by new images. Image-based classification enables the system to give reasons for selecting matched cases and solutions to the problem description. Therefore this hybrid system has the advantages of deep learning (DL) and CBR and benefits from both. DL helps researchers absolutely to treat and detect diseases by analyzing medical data (e.g., medical images). One of the representative models among the various deep-learning models is a convolutional neural network (CNN). This classification of skin cancer by CNN as an image-based classification is comparable to the dermatologists’ detection [ ].