Application of deep convolutional neural network in multiclass skin cancer classification using custom CNN architecture

Abstract

Skin diseases represent a significant global health concern, and prompt and pre- cise diagnosis is necessary for efficient treatment. Convolutional Neural Networks (CNNs), in particular, have shown tremendous promise in the diagnosis of skin dis- eases due to their capacity for processing and learning from complex patterns in visual data. Employing 28x28 RGB images taken from the HAM10000 dataset, the purpose of this work is to develop and assess a customized CNN model created exclusively to aid in the classification of different skin conditions. This method al- lows the model to efficiently learn the distinctive characteristics of each type. Our model is evaluated using a number of metrics, such as accuracy, precision, recall, and F1-score. We have also compared our results to well-known pre-trained models like ResNet50 and EfficientNetB0/B2. In comparison to existing pre-trained mod- els, our own model performs better due to its increased test accuracy, reduced test loss, and computational parameters. Additionally, it has fewer trainable parame- ters as well as a shorter training time per epoch, which makes it appropriate for deployment in situations with constrained computational resources. In conclusion, Our model promises to improve diagnostic accuracy, perhaps enabling earlier and more effective methods for diseases of the skin because of its higher performance and computational advantages.