Low-cost image analysis with convolutional neural network for herpes zoster

Herpes zoster virus (HZV) or varicella-zoster virus (VZV) affects the trigeminal nerve, at the earliest possible stage will avoid the eyes injuries. In this paper, the new framework develops a new method with convolutional neural networks (CNN), the detection for the early stage of the HZV is tested with 1,000 images. It is 89.6% with low-cost image analysis, besides, the database has been analyzed with other architectures in order to validate the most appropriate algorithm. The process is pre-processing, segmentation, extraction, and classification. The VZV produces two illness: i) Varicella called chickenpox, and ii) Herpes Zoster. In order to obtain a machine learning process, it considers building blocks of convolutional layer neural network associated to a new process for early Herpes Zoster (HZ) disease detection system, structured in four stages as pre-processing, segmentation, extraction and classification. In particular, the new process includes a classification process with a comparison between the K-Nearest Neighborhood (KNN), artificial neural networks (ANN), and logistic model tree (LMT) regression for the comparison. The effectiveness during eight days is 98.1%, for early detection with minimal information. However, the training process produces 33% false positives and an average 90% true positive rate. Early HZ detection and the failures associated with electronic devices were shown and used for facial and pattern recognition associated with nerve location. With this research, the difficulties concerning to the data management and deep learning were corroborated during eight days of the illness, to better understand the process and technology that enable a successful classification.