Hybrid Algorithm Using CNN and LSTM to Improve Performance in Skin Lesion Image Classification

Skin lesions are a condition of abnormal growths or changes in the appearance of the skin, such as lumps, abrasions, cysts, or color differences. Skin lesions can be detected as cancerous and non-cancerous. Detecting earlier skin lesions can be attained through various methods, featuring the use of dermoscopic tools with the assistance of a dermatologist. Dermoscopy has enhanced accuracy in diagnosing the early stages of skin cancer. However, the persistent challenge of relatively low accuracy rates still occurs. Although a dermatoscopy is used, the identification of skin cancer remains imperfect. This research proposes a deep learning approach employing a hybrid algorithm that combines Convolutional Neural Networks and Long ShortTerm Memory networks. This method aims to enhance the efficacy of early skin lesion detection and classification. The dataset employed in this study comprised 2,736 image samples from a combination of two datasets, PAD-UFES-20 and ISIC dataset. These samples are categorized into six distinct classes. The research findings show an improvement in accuracy, increasing from 96.27% with the CNN network to $97.26 \%$ after integrating the CNN-LSTM network. In addition, the integrated CNN-LSTM network model demonstrated $91.83 \%$ sensitivity, $91.83 \%$ precision, and $91.82 \%$ F1-Score. This finding surpassed the performance of the CNN network, which achieved lower metrics with $88.83 \%$ sensitivity, $88.83 \%$ precision, and $88.82 \%$ F1-Score. This research contributes to the early detection of skin diseases through feature extraction techniques, which can be the basis for developing a more accurate and better skin lesion classification system.