Skin Lesion Analysis using Graph Convolutional Networks and Enhanced Secretary Bird Optimization

Abstract

An essential feature of automatic dermatology diagnosis is skin lesion analysis, especially for the early diagnosis of melanomas and other skin malignancies. This study suggests a hybrid system that combines an improved secretary bird optimization algorithm (E-SBOA) for reliable multilevel classification of images with graph convolutional networks (GCNs) for advanced representations of features. Under varying illumination and texture circumstances, our method enhanced feature extraction, reduce imbalances in classes, and increase the accuracy of segmentation. Considerable performance gains over conventional CNN and meta-heuristic techniques are shown by extensive testing on publically accessible datasets.

Keywords