Deep Learning Approaches for Pneumonia Classification in Healthcare

Abstract

In the past two decades, there has been a sharp rise in the use of deep learning for medical image processing and analysis. Recent challenges, for instance, the most well-known ImageNet Computer Vision competition, have almost entirely incorporated deep learning approaches for providing the best result. The concept of Image classification was later extended to Image Segmentation and Object Detection which proved to perform extremely well using state-of-the-art classification algorithms as their backbone architecture. The accuracy of the algorithm and approach has a significant impact on the medical field as there is a constant need for accurate and computationally efficient models. The existing object detection and segmentation approaches need large data for providing accurate results, unlike classification algorithms in which accuracy can be achieved with a relatively smaller amount of data. Hence, for the overall increase of model accuracy, there is a need for image augmentation to be incorporated. In this paper, several deep learning methodologies such as classification, object detection, ensemble, and segmentation for pneumonia classification and detection have been reviewed and an ensemble-based approach for the classification of Pneumonia using chest X-rays has been proposed.