Cost-Effective and Transparent Diagnosis of COVID-19 and Pneumonia using X-ray Images: A Machine Learning Approach

Abstract

The demand for rapid, cost-effective, and precise diagnostic instruments for chest-related conditions, including COVID-19 and pneumonia, has become more pressing in the aftermath of the COVID-19 pandemic. Despite the remarkable results that deep learning-based approaches have achieved, they frequently necessitate large annotated datasets, substantial computational resources, and a lack of interpretability. In contrast, this study suggests a low-cost hybrid machine learning-based diagnostic framework that classifies cases into three categories: Normal, Pneumonia, and COVID-19, based on interpretable features extracted from chest X-ray images. In order to quantify structural complexity, texture variability, and edge information, three critical features—Fractal Dimension, Entropy, and Edge Density—were extracted from each image. The statistical analysis of ANOVA confirmed that all three features exhibited significant variation among the three classes (p < 0.005). This was further substantiated by the visual separability demonstrated by boxplots. The machine learning classifier was trained using the extracted features, resulting in an overall accuracy of 77.13 percent. The model was notably effective in distinguishing between Pneumonia (F1- score: 0.84) and Normal (F1-score: 0.64). However, the classification of COVID-19 (F1-score: 0.53) exhibited some overlap with Pneumonia. The ROC curves demonstrated a robust discriminative capacity, with AUC values of 0.87 for Normal, 0.84 for COVID-19, and 0.82 for Pneumonia. This method is interpretable and lightweight, and it offers a cost- effective solution for the preliminary diagnosis of pulmonary disease. Additionally, it establishes a solid foundation for future improvements that will incorporate more complex features and models.