Machine Learning and Deep Learning Methods for Pneumonia Detection: A Systematic Review

Abstract

Pneumonia is a very frequent respiratory infection that remains a serious issue in all health services. Routine and accurate diagnosis is crucial to limit the severity of the disease and benefit treatment. In the past few years, image analysis algorithms using artificial intelligence have received significant research interest as a method for the identification of pneumonia in medical images. The ability of machine learning and deep learning to detect patterns of disease in chest X-ray and computed tomography (CT) scans has been demonstrated to be useful. This systematic review offers a comprehensive assessment of the current research on machine learning and deep learning methods for the detection of pneumonia. This paper compares the various approaches, data sets, image preprocessing techniques and assessment methods reported in the literature. Popular machine learning algorithms are evaluated, along with contemporary deep learning architectures like Convolutional Neural Networks, ResNet, DenseNet, EfficientNet and Transformer-based architectures. The review also examines the datasets which are frequently used for model training and evaluation. The analysis shows that overall the performance of predictive deep learning methods is higher than the predictive performance of conventional machine learning methods as the deep learning methods are able to learn complex image features automatically. However, problems concerning data quality, class imbalance, model interpretability and deployment in the real world still persist. Current limitations in the technology are outlined and new research directions to aid in developing more reliable and clinically relevant pneumonia detection systems are discussed.