Classification of Melanoma Skin Cancer Based on Image Data Set Using Different Neural Networks

Abstract

Melanoma is an aggressive skin cancer type that is incredibly terrifying because of its ten dency to become prevalent across the body if not detected and treated on time. In the field of medical image diagnosis, computer vision can play a significant role, as shown by several existing technologies. In this paper, we present neural network models such as basic CNN, ResNet-18, and EfficientNet-B0 for image processing in melanoma skin cancer detection. This data set has binary classes such as benign and malignant with a 10605 sample size, where 9605 images for training, and 1000 images for testing the model’s performance. The process of segmentation, extracting features, classification, and pre-processing process are the processes that were used in this study. A classification with a success of 97% accuracy was produced by EfficientNet-B0, which outperformed 87% from ResNet-18 and 80% from CNN for the classification of malignant and benign. According to other evaluation perfor mances such as sensitivity, specificity, f1-score, precision, error rate, Mathew’s correlation coefficient, geometric mean, and bookmaker informedness, EfficientNet-B0 outperforms ResNet-18 and CNN. The findings of this research indicate Neural Network models specif ically EfficientNet-B0 show significant potential for accurate and efficient melanoma skin cancer detection to save lives.