Detection of Outer Throat Infection using Deep Convolutional Neural Network

Abstract

It is integral for physicians to be able to assess through a thorough history and physical exam. However, it has been increasingly difficult to perform rigorous physical examinations because of the COVID-19 pandemic. Thus, there is an increasing relevance of improved techniques of assessment through image classification using Deep Convolutional Neural Network. The ResNet50 architecture will be used as a classifier in Convolutional Neural Network. This type of network subtracts the feature learned from any given layer for which the ResNet50 learns by utilizing the found shortcut connections, which proved to be easier compared to some types of Convolutional Neural Networks. The learned features from ResNet50 are essential to Fully Connected Layers in Neural Networks as it aids the neural network to decide based on the features extracted and come up with a result using softmax function. The researchers are able to train a network and test it. It is very convenient for a patient, especially in the midst of the COVID-19 pandemic, to be assessed without having to be physically examined by a physician. In the GUI, the patient must register on the web app and take a photo of the throat and send it – the patient will reserve a notification containing the diagnosis of the photo. The network obtained positive results, with a 92% accuracy rate in looking for healthy, inflamed throat, inflamed throat and swollen tonsilitis, inflamed throat/ swollen tonsils, and white spots in throat images.