Pneumonia Radiographic Classification Using Shallow Learning Approach

Pneumonia is a common lung infection usually suffered by children under the age of 5 years old. Pneumonia cases occur as a result of air pollution, which is particularly prevalent in developing countries where the World Health Organization (WHO) estimated that there will be more than 4 million casualties caused by pneumonia. Hence, the use of technologies is one of the ways that can effectively improve existing clinical approaches to pneumonia cases. This paper focuses on the implementation of a Shallow Learning (SL) technique to classify normal, viral, and bacterial pneumonia cases from chest radiographic images. To develop the system, Gray-Level Co-Occurrence Matrices (GLCM) texture features extracted the chest radiographic images into four (4) types of features namely energy, contrast, correlation, and homogeneity. Classification features were extracted from a radiographic image using K-Nearest Neighbours (KNN) where three classes were produced by this supervised learning. The finding reveals that the implementation of SL approach, which is KNN, was able to achieve a high accuracy of 86.67%. It shows that KNN can be used to classify the lung radiographic images without utilizing any complex learning approach.