A Deep Learning Approach to Automate Classification of Arsenic-Affected Skin using EfficientNet-B1

Arsenic contamination in groundwater is a major health hazard, particularly in areas where it causes chronic arsenic poisoning, which is manifested in skin lesions and other dermatological conditions. Early and accurate detection of arsenic-affected skin is important for timely medical intervention. This paper proposes the automatic classification system using EfficientNet-B1, which is one of the most outstanding deep learning models for detecting arsenic-induced skin lesions using clinical images, in the model, training, and testing of a dataset comprising healthy and arsenic-affected skin, with augmentation techniques employed to increase robustness and generalizability from the image dataset. Several deep earning models were compared for this classification task: VGG-19, ResNet-50, MobileNetv2, EfficientNet-B0, and EfficientNetB1. EfficientNet-B1, after fine-tuning, achieved the highest performance of 94.69% accuracy and proved outstanding in this application. These results point out that EfficientNet-B1 is a reliable and scalable tool for automatic diagnosis and can thus assist healthcare providers in resource-constrained settings. Future work will include an extension of the dataset, improving model interpretability, and exploring clinical practical use of the system.