Using Convolutional Neural Networks for Enhanced Pneumonia Detection via Chest X-Rays
Waqar Babar, Raja Hashim Ali, Alishba Faheem, Syed Ahmed Mansoor
Abstract
As a deadly lung disease, pneumonia remains a leading cause of child mortality under five years old. Machine learning, especially deep learning have played a significant role in the improved performance for detection and identification of various diseases in the field of healthcare. Neural networks, especially the recent development of newer architectures, have revolutionized the object identification, and classification applications in clinical diagnosis of various diseases. This study presents the application of Convolutional Neural Networks (CNN) for the timely and accurate detection of pneumonia using chest X-rays, a development with considerable potential in aiding clinical diagnosis. The research deployed dropout regularization in model design to mitigate overfitting and relied on recall and F1 scores for thorough model evaluation. Although comparable studies achieved higher overall accuracy, our models registered a recall rate of 98%, crucial in reducing false negatives and enhancing patient safety. This suggests the potential of our CNN model as a vital tool for healthcare professionals in early pneumonia detection in children and adults, with the capacity to process a high volume of X-ray images rapidly and accurately. The successful model construction was enabled through various parameter tuning techniques, thus enhancing patient care efficiency and the potential to decrease mortality rates.