CLASSIFICATION MODEL FOR COVID-19 AND PULMONARY (TB) FROM X-RAY IMAGES USING HOG-PCA-LEARNING ALGORITHMS

Abstract

CoronaVirus Disease 2019 (COVID-19) is induced by a new virus SARS-CoV2. Its incidence is
unprecedented and caused a huge dent to the health care system and the whole world. The hasty spread
of COVID-19 and lack of fast diagnosis drove machine learning researchers to build intelligent
response system to help the healthcare delivery personnel to manage the disease and the patient. The
aim of this study is to build a COVID-19/ Pulmonary Tuberculosis (PTB) classification model from
Chest X-ray (CXR) images. Due to small sample size of COVID-19 CXR image available, a fourphased
method is adopted involving feature extraction, selection, modelling and classification.
The CXR images of lungs infected with COVID-19, PTB and Normal were obtained from databases.
Features were extracted from these images by Histogram of Oriented Gradient (HOG)descriptor.
Principal Component Analysis (PCA) technique was used to extract the most relevant features to
enhance classification. For this study, from 1327 CXR image samples 46,657 features were extracted.
But 675 relevant and important features were selected with 95% explained variance of PCA. A
number of learning algorithms such as Support Vector Machine (SVM), k-Nearest Neighbor (k-NN),
Random Forest (RF), eXtreme Gradient B boosting (XGBoost) and Decision Tree (DT)classifiers
was used and evaluated. The experimental results obtained showed that SVM classifier produced best
results of 0.97 based on precision, accuracy, F1-Score and recall metrics when compared to other
learning algorithms.