Detection of Lung Diseases for Pneumonia, Tuberculosis, and COVID-19 with Artificial Intelligence Tools

Chest X-ray imaging is a low-cost, easy way to diagnose lung abnormalities caused by infectious diseases such as COVID-19, pneumonia, or tuberculosis. The primary objective of the study is to carefully analyse and evaluate several classification strategies to determine which technique based on machine learning or deep learning would be more useful for detecting lung infectious illness using chest X-rays of three pulmonary infectious diseases: pneumonia, TB, and COVID-19. To notify physicians and radiologists of probable aberrant results, the performance of numerous classifiers—deep learning algorithms (CNN) and conventional machine learning algorithms—for distinguishing between normal and pathological chest radiographs was assessed and compared. The comparative analysis is based on three important criteria: the performance metrics (precision, accuracy, recall, and f1-score), minimising overfitting, and reducing false negative and false positive counts. Results of evaluation show convolutional neural network model accuracy across training and test samples was 94.71% and 90.22% for dataset I, 96.31% and 95.60% for dataset II, and 99.01% and 99.04% for dataset III, respectively, which is better than the conventional ML models. The experimental results in this paper also show that a deep learning framework such as CNN outperforms traditional machine learning approaches, viz., support vector machines, logistic regression, k-nearest neighbours, Naive Bayes, decision trees, and random forests on large X-ray image datasets, as it also shows better results for precision, F1 score, and recall, minimum overfitting, and a reduced number of false negative and false positive counts for pneumonia, TB, and COVID-19 lung diseases.