A three-tier BERT based transformer framework for detecting and classifying skin cancer with HSCGS algorithm

• Published in: Multimedia tools and applications Vol. 83; no. 17; pp. 51441 - 51467
• Main Authors: George, Joseph, Rao, Anne Koteswara
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.05.2024; Springer Nature B.V
• Subjects: Accuracy; Algorithms; Cancer; Classification; Computer Communication Networks; Computer Science; Data Structures and Information Theory; Feature extraction; Graph neural networks; Image contrast; Image enhancement; Machine learning; Medical imaging; Multimedia Information Systems; Optimization; Optimization algorithms; Skin cancer; Special Purpose and Application-Based Systems; Transformers; Gated recurrent unit; Skin cancer detection; Chimp optimization algorithm; BiLSTM; Glow swarm optimization
• ISSN: 1573-7721; 1380-7501; 1573-7721
• DOI: 10.1007/s11042-023-17590-1

Skin cancer is the process of identifying and diagnosing, a disease in which abnormal skin cells grow and spread uncontrollably. An innovative deep learning-based skin cancer detection model is introduced in this research work. The proposed model is divided into five main phases: (a) Pre-Processing (b) Segmentation (c) Feature Extraction (d) 3-Tier Classification (e) post-processing. Initially, the collected raw image is pre-processed via contrast enhancement, decimal scaling, and augmentation methods. From the pre-processed image, the important feature is extracted by using the statistical features like mean, variance, and information gain. Then, from the extracted image, the region of interest ROI is identified via fuzzy assisted Kapur’s multi-level thresholding. The optimal features are selected using the hybrid Self-Improved Chimp Optimization algorithm with Glow Swarm Optimization algorithm (HSCGS). The three-tier classification using the BERT based Transformer with HSCGS based Gated Recurred Unit (GRU), BiLSTM, and Graph Neural Network is projected for classification. The proposed model is implemented using the PYTHON platform. The findings are evaluated in terms of accuracy, sensitivity, precision, FPR, FNR, etc. using the present models. The proposed model has recorded the highest detection accuracy as 97% and highest MCC and NPV values. Proposed model has shown the best performance and has outperformed other models.