Detection of melanoma with hybrid learning method by removing hair from dermoscopic images using image processing techniques and wavelet transform

Abstract

Melanoma is one of the most dangerous types of skin cancer. Detecting melanoma is complicated and takes too much time, which doctors and healthcare professionals accomplish with the help of dermoscopic images. In the last few years, artificial intelligence-based (AI) systems have emerged to detect melanoma and save healthcare workers' time. In addition, with AI-based systems, the need for an expert is eliminated, and diagnoses can be made with a high accuracy rate. Thanks to these systems, the workload of healthcare professionals will be reduced, and patients will gain time on the way to early diagnosis and treatment. This paper proposes the detection of melanoma with hybrid learning techniques. In the system, first of all, using image processing techniques (masking for saturation and wavelet transform) removes obstacles such as hair, air bubbles, and noise in dermoscopic images to increase the algorithm's detection speed. Another critical point of this process is to make the lesion more prominent for detection. As an AI block, an original hybrid model combining deep learning and machine learning has been used for the first time in melanoma detection. After stabilization, the Hamm 1000 (ISIC 2018) and ISIC 2020 datasets were used to measure the performance ratio of the designed system. The best accuracy score in the literature is obtained at 99.44% and 100%, thanks to these two methods.