Detection of Thoracic Diseases Using Chest X-Ray: A Comparative Study of Binary Class and Multiclass Classification Using Deep Learning

Identifying thoracic ailments stands pivotal in medical imaging, offering insights for disease recognition and treatment. Chest X-ray (CXR) imaging emerges as a prevalent non-invasive and cost-effective technique for thoracic disease detection [1–4]. This research delves into a comparative exploration between binary class and multiclass classification methodologies in identifying thoracic diseases through CXR images. Our approach involves training and assessing deep learning models leveraging a publicly accessible CXR image dataset, juxtaposing the performance of binary and multiclass models using diverse performance metrics. We formulated 17 distinct binary class models and orchestrated their ensemble, alongside crafting a solitary multiclass classification model [25–26]. The assessment of these models encompassed various performance metrics, including F1 score, accuracy, precision, and recall. The outcomes advocate for binary classification's pragmatic suitability, attributed to its heightened accuracy and simplified implementation. This inquiry strives to refine the precision and efficacy of CXR image analysis in clinical practice.