SCED: A Multi-criteria-based Modeling of Deep Learning Architectures for Skin Cancer Analysis

AI is progressing in the healthcare sector to provide solutions to challenging problems these days. Many healthcare systems expect to be precise, early disease diagnosis. However, in the case of skin cancer disease, most of the existing systems are neither precise nor optimal. In this regard, this work proposes a novel integrated approach SCED - Skin Cancer Early Diagnosis system. The proposed work is three-fold. First, to investigate the state-of-art classification architectures such as EfficientNet integration with CNN, INN, and heuristic search based optimal parameter search. Second, to design an optimal architecture to not only classify the disease type, also perform early disease diagnosis with growth estimation computation application. Third, to provide a severity estimation from multi-criteria-based decision modeling and expert systems. Significantly, with the proposed approach SCED, the proposed heuristic search Involution block achieved 84.7% test accuracy. Further, the EfficientNet model is explored with CNN and a novel architecture is proposed with Involution blocks. Notably, the investigated INN architecture not only reduces overall parameters by 28.6% but also reduces overall training time by 45% per epoch. On other hand, the novel EfficientNet-INN provided 93.71% test accuracy and 95.99% for EfficientNet-B0, reducing overall learning parameters during training by 92.59% which is essential for any real-time deployment and validation.