Detection and Classification of Lung Disease Using Deep Learning Architecture from X-ray Images

Abstract

The chest X-ray is among the most widely used diagnostic imaging for diagnosing many lung and bone-related diseases. Recent advances in deep learning have shown many good performances in disease identification from chest X-rays. But stability and class imbalance are yet to be addressed. In this study, we proposed a CX-Ultranet (Chest X-ray Ultranet) to classify and identify thirteen thoracic lung diseases from chest X-rays by utilizing a multiclass cross-entropy loss function on a compound scaling framework using EfficientNet as a baseline. The CX-Ultra net achieves 88% average prediction accuracy on NIH Chest X-ray Dataset. It takes ≈ 30% less time than pre-existing state-of-the-art models. The proposed CX-Ultra net gives higher average accuracy and efficiently handles the class imbalance issue. The training time in terms of Floating-Point Operations Per Second is significantly less, thus setting a new threshold in disease diagnosis from chest X-rays.