An Evaluation of the Impact of Data Augmentation and Loss Functions on Disease Recognition in Chest X-ray Images

Abstract

X-ray examinations are frequently utilized in health check-ups due to their short duration and low radiation exposure. However, detecting diseases through X-ray imaging can be challenging. Specialized physicians are required to accurately diagnose these diseases, but their numbers are insufficient relative to the workload. This study aims to enhance the accuracy of disease identification models for diagnostic support by utilizing a long-tail chest X-ray image dataset. The research investigates the effectiveness of data augmentation and the optimization of two loss functions and their weight adjustments.