A Rest API to Classify Pneumonia Infection From Chest X-ray Images Using Multi-Layer Perceptron and LeNet

Abstract

Pneumonia remains the most common reason for inpatient stays and fatalities among adults and children in the world. It became worse during Covid 19 pandemic. Most African countries like South Africa were and are still seriously affected. The situation is worse in rural areas because of several reasons, among them; not having enough X-rays machines, having no or few radiologists to analyze and interpret the X-ray pictures to determine if the pictures are normal pictures or pneumonia. The ability to accurately classify these two types of pneumonia can guarantee effective treatment which will boost survival chances. Artificial Intelligence (AI) is a cost-effective approach and can play a pivotal role in easily analyzing and interpreting X-ray images. This research used CRoss Industry Standard Process for Data Mining methodology in developing a simple Rest API model that would classify the chest X-ray image if it were normal, the person has pneumonia caused by bacteria or virus. Multi-Layer Perceptron (MLP) model had a training accuracy of 73.89%, validation accuracy of 75.46%, and test accuracy of 75.46% whereas LeNet had 78.49%, 76.51%, and 76,51%, respectively. This study demonstrated to the public that AI models may be developed to aid health professionals in the early diagnosis, classification, analysis, and interpretation of X-ray images for pneumonia. In the future, the model created should convert the English interpretations into South African local languages like isiXhosa, Zulu, Venda, and many others. Thus, making it easier for the local communities to understand giving them a sense of belonging.