Impact of Gender and Chest X-Ray View Imbalance in Pneumonia Classification Using Deep Learning

The imbalance of gender and age in Chest X-rays (CXRs) datasets for pulmonary diseases and COVID-19 screening were found to affect deep learning models’ performance. However, the optimal training requirements for each gender and the best use-cases of CXR views are yet to be discovered. Our objective is to determine the impact of the view (PA or AP) and the gender (Male or Female) on the performance of deep learning models in the classification of pneumonia using CXRs and deduce the optimal training dataset specifications. We realize that CXRs that are for female patients are better tested on models trained with female Anterior-Posterior CXRs, and male patients are better tested on models trained with male Posterior-Anterior CXRs. Moreover, the Anterior-Posterior CXRs are preferred to train the model if both genders are present. the drawn up conclusions will be a valuable asset to an optimized pulmonary diseases deep learning classifier.