Deep learning in skin lesion analysis for malignant melanoma cancer identification

• Published in: Multimedia tools and applications Vol. 83; no. 6; pp. 17833 - 17853
• Main Authors: Sivakumar, M. Senthil, Leo, L. Megalan, Gurumekala, T., Sindhu, V., Priyadharshini, A. Saraswathi
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.02.2024; Springer Nature B.V
• Subjects: Accuracy; Applications programs; Artificial neural networks; Automation; Cancer; Classification; Computer Communication Networks; Computer Science; Data collection; Data Structures and Information Theory; Deep learning; Diagnosis; Feature extraction; Image classification; Lesions; Machine learning; Medical imaging; Melanoma; Model accuracy; Multimedia Information Systems; Neural networks; Preprocessing; Skin cancer; Skin diseases; Special Purpose and Application-Based Systems; Track 2: Medical Applications of Multimedia; Artificial neural network; Malignant; ResNet50; Convolutional neural network; Machine learning; Benign
• ISSN: 1573-7721; 1380-7501; 1573-7721
• DOI: 10.1007/s11042-023-16273-1

The higher rate of skin diseases caused by infections, allergies, lifestyle changes, increased use of chemicals, unhealthy food habits, and hereditary reactions, among other factors demand effective diagnostic mechanisms. These mechanisms should facilitate and expedite the precise diagnosis of malignant melanoma. In addition, the existing technology-assisted tests require elaborate procedures to accurately diagnose the skin disease to be treated. The deep learning-enhanced inspection involves neural networks that diagnose skin diseases by analyzing data for skin lesions such as rashes, boils, and cancer growth. This paper proposes an automated high-precision diagnostic model with the web application to identify Malignant Melanoma Cancer. The diagnostic model framework employs a convolutional neural network (CNN) with ResNet50 for data collection, preprocessing, segmentation, enhancement, feature extraction, classification, and malignant melanoma identification to improve accuracy. The proposed model eliminates unwanted noise, enhances spectral image information, and improves classification accuracy through preprocessing and mixed hybrid pooling phases. Performance analysis shows that the proposed Malignant Melanoma Cancer detection model achieves the highest accuracy of 94% and an F1-score of 93.9%. Results from a training and testing dataset of images from the International Skin Image Collaboration (ISIC) demonstrate a significant improvement over traditional methods. Further, the web application developed for this model accelerates the precise diagnosis of Malignant Melanoma Cancer with accurate information compared to other methods in practice. The potential of applying the proposed skin cancer classification model automates the malignant and benign image classification process, speeds up diagnosis and treatment, and reduces the risk of misdiagnosis.