An enhanced skin lesion detection and classification model using hybrid convolution-based ensemble learning model

One of the major health concerns that affect the well-being of people around the world is called as the skin cancer. Among diverse kinds of skin cancer, melanoma is regarded as the most harmful disease along with a higher rate of mortality. Earlier detection as well as the screening process is considered the more complex job for dermatologists due to the enormous variations in morphological attributes of skin cancer. Therefore, there is a requirement for the most reliable and efficient diagnosis system, which has aided dermatologists in adequate decision-making and diagnosis.

An advanced deep learning technique-based skin lesions classification and detection model for early detection and efficient classification of skin lesions is proposed.

The dermoscopic images are collected from online sources in the earlier stage. Then, the collected images are segmented with the help of dilated Mask-Regions with Convolutional Neural Networks (RCNN) with an attention mechanism for segmenting the abnormal regions. After that, the segmented images are classified using adaptive hybrid convolution-based ensemble learning (AHC-EL), which is used along with techniques like residual attention network (RAN), MobileNet, and Inception. Here, parameters optimization takes place using a hybrid optimization algorithm namely, fitness-aided battle royale and red deer algorithm (FBR-RDA) for enhancing the classification performance. Finally, the classified outputs of skin lesion classification are obtained based on high ranking between the ensemble learning techniques.

Experimental analysis is carried out between proposed and conventional approaches to verify the efficacy of the recommended method.