Effects of objects and image quality on melanoma classification using Spatio Temporal Joint graph Convolutional Network

•Malignant melanoma has high mortality; early diagnosis is crucial.•IEMBF pre-processing removes noise and prepares images for analysis.•STJGCN model classifies melanoma as benign or malignant.•RPOA fine-tunes STJGCN parameters for optimal detection accuracy.•EOIQ-MC-STJGCN achieves 97.86% accuracy in melanoma detection. Malignant melanoma, a highly lethal subtype of skin cancer, is categorized by the abnormal growth of melanocyte cells and is associated with elevated mortality rates. Improving the prognosis of patients with melanoma requires an early and precise diagnosis. Recently, deep learning models have emerged as dominant tools in computer-aided diagnosis (CAD) systems for the categorization of potential melanoma lesions. In clinical settings, obtaining high quality skin images can be challenging due to blurry, noisy, low-contrast images that may contain extraneous objects such as rulers or hair. To overcome these challenges, the Spatio Temporal Joint Graph Convolutional Network (STJGCN) was utilized to detect and classify melanoma from dermoscopy images. The input images are collected from the 2020 Challenge Dataset of the International Skin Imaging Collaboration (ISIC). The input images are provided to Information Exchange Multi-Bernoulli-Filter (IEMBF) based pre-processing for hair noise removal, resizing, random rotations, cropping, and splitting from input dataset. The pre-processed images are provided to Spatio Temporal Joint Graph Convolutional Network (STJGCN) for detecting and classifying melanoma as benign and malignant. The Red Panda Optimization Algorithm (RPOA) is used to fine-tune STJGCN parameters. The proposed EOIQ-MC-STJGCN approach is implemented in Python and various performance metrics are analyzed to assess the efficacy of the proposed approach. The proposed approach attains the highest accuracy of 97.86% in melanoma detection and classification compared to various existing approaches. By demonstrating the efficiency of the proposed EOIQ-MC-STJGCN technique, this research contributes to faster and more accurate melanoma diagnosis, ultimately facilitating earlier intervention and treatment.