Data augmentation using Fast and Adaptive Bidimensional Empirical Mode Decomposition for pulmonary tuberculosis diagnosis in chest X-rays

Abstract. Automation in medical diagnostics based on machine learning algorithms relies heavily on the quality and volume of training data. For pulmonary tuberculosis diagnosis in chest X-rays image quality varies due to differences in equipment and acquisition conditions, and the availability of high-quality data is limited due to legal constraints and the smaller size of public datasets compared to those for some other lung diseases. Additionally, concerns regarding cross-dataset compatibility and discrepancies between training and target data distributions further complicate the analysis. To mitigate these issues, we propose a data augmentation technique utilizing the Fast and Adaptive Bidimensional Empirical Mode Decomposition (FABEMD) algorithm. Experiments have demonstrated its effectiveness for pulmonary tuberculosis diagnosis in chest X-rays.