Pulmonary Tuberculosis Screening With an Optimized Deep Learning Method for Varying Image Resolutions

There may be a possibility of error in the manual diagnosis of tuberculosis using chest X-ray images. To effectively diagnose TB from chest X-ray images, researchers have been working hard to establish a computerized decision support system. For the healthcare industry to advance, any advances that enhance diagnostic procedures while retaining quality and safety are essential. This paper proposes an improved deep learning-based model that picks the classifier’s hyperparameters and extracts features from X-ray images. The goals of this study are to improve accuracy and reduce the amount of false positive and false negative results. In this study, the diagnostic performance for detecting tuberculosis is analyzed through deep learning approach optimized with selected hyper-parameters. In addition to performance optimization, this study also analyses the impact of image sizes on model’s overall performance. Here deep learning model (CNN) performance is evaluated for three different sizes of images viz. 150 × 150, 248 × 248, and 500 × 500 and it has been observed that with selected hyperparameters through keras tuner, CNN has outperformed for image size 248 × 248 compared to image sizes 150 × 150, and 500 × 500. Therefore, in addition to parameter tuning, model performance also depends on the size of the images.