Skin cancer detection using ensemble of machine learning and deep learning techniques

• Published in: Multimedia tools and applications Vol. 82; no. 18; pp. 27501 - 27524
• Main Authors: Tembhurne, Jitendra V., Hebbar, Nachiketa, Patil, Hemprasad Y., Diwan, Tausif
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.07.2023; Springer Nature B.V
• Subjects: Automation; Cancer; Computer Communication Networks; Computer Science; Data Structures and Information Theory; Datasets; Deep learning; Dermatology; Feature extraction; Image classification; Machine learning; Medical imaging; Melanoma; Multimedia Information Systems; Neural networks; Skin cancer; Special Purpose and Application-Based Systems; Deep learning; Contourlet transform; Melanoma; Machine learning; Skin cancer
• ISSN: 1380-7501; 1573-7721
• DOI: 10.1007/s11042-023-14697-3

Skin cancer is one of the most common forms of cancer, which makes it pertinent to be able to diagnose it accurately. In particular, melanoma is a form of skin cancer that is fatal and accounts for 6 of every 7-skin cancer related death. Moreover, in hospitals where dermatologists have to diagnose multiple cases of skin cancer, there are high possibilities of false negatives in diagnosis. To avoid such incidents, there here has been exhaustive research conducted by the research community all over the world to build highly accurate automated tools for skin cancer detection. In this paper, we introduce a novel approach of combining machine learning and deep learning techniques to solve the problem of skin cancer detection. The deep learning model uses state-of-the-art neural networks to extract features from images whereas the machine learning model processes image features which are obtained after performing the techniques such as Contourlet Transform and Local Binary Pattern Histogram. Meaningful feature extraction is crucial for any image classification roblem. As a result, by combining the manual and automated features, our designed model achieves a higher accuracy of 93% with an individual recall score of 99.7% and 86% for the benign and malignant forms of cancer, respectively. We benchmarked the model on publicly available Kaggle dataset containing processed images from ISIC Archive dataset. The proposed ensemble outperforms both expert dermatologists as well as other state-of-the-art deep learning and machine learning methods. Thus, this novel method can be of high assistance to dermatologists to help prevent any misdiagnosis.