Automatic Monkeypox Disease Detection from Preprocessed Images using MobileNetV2
Ramya Mohan, Robertas Damaševičius, David Taniar, N. Sri Madhava Raja, V. Rajinikanth
Abstract
The prevalence of infectious diseases in humankind is increasing globally due to a variety of reasons, and accurate diagnosis and treatment will help control/cure the disease. A severe health problem is associated with Monkeypox (Mpox), a communicable illness caused by the monkeypox virus. This research aims to develop a computerized tool to detect Mpox from pre-processed images using the pre-trained lightweight deep-learning scheme (PLDS). This tool consists of the following phases; (i) Image collection and tri-level thresholding based pre-processing, (ii) Feature extraction using selected PLDS, and (iii) five-fold cross-validation supported binary classification. As part of this research, we examine the possibilities for developing an accurate Mpox detection system using PLDS, including; (i) Kapur's thresholding, (ii) chosen optimizers, and (ii) chosen activation functions. This experimental investigation utilizes augmented images from the Monkeypox Skin Images Dataset (MSID), and the developed tool with MobileNetV2 achieves 98.7% detection accuracy when Kapur's thresholding is applied. Further, this tool presents a testing accuracy of >90% on the original MSID images, confirming the proposed research's significance.