In-Line Image Transformations for Imbalanced, Multiclass Computer Vision Classification of Lung Chest X-Rays

TL;DR

This paper introduces in-line image transformations to address class imbalance in lung X-ray datasets, enabling a CNN to classify multiple lung conditions, including COVID-19, with high accuracy.

Contribution

It proposes a novel image transformation technique to balance imbalanced datasets for improved multiclass lung X-ray classification.

Findings

Achieved 94% accuracy in multiclass classification

Demonstrated effectiveness of image transformations for data balancing

Applied simple CNN architecture successfully

Abstract

Artificial intelligence (AI) is disrupting the medical field as advances in modern technology allow common household computers to learn anatomical and pathological features that distinguish between healthy and disease with the accuracy of highly specialized, trained physicians. Computer vision AI applications use medical imaging, such as lung chest X-Rays (LCXRs), to facilitate diagnoses by providing second-opinions in addition to a physician's or radiologist's interpretation. Considering the advent of the current Coronavirus disease (COVID-19) pandemic, LCXRs may provide rapid insights to indirectly aid in infection containment, however generating a reliably labeled image dataset for a novel disease is not an easy feat, nor is it of highest priority when combating a global pandemic. Deep learning techniques such as convolutional neural networks (CNNs) are able to select features that distinguish between healthy and disease states for other lung pathologies; this study aims to leverage that body of literature in order to apply image transformations that would serve to balance the lack of COVID-19 LCXR data. Furthermore, this study utilizes a simple CNN architecture for high-performance multiclass LCXR classification at 94 percent accuracy.