Deep Learning-Based Multi-label Image Classification for Chest X-Rays

Radiology employs radiation and imaging technologies, such as chest X-ray images, to diagnose and treat disorders. Chest X-rays (CXRs) are useful in emergency diagnosis and treatment; however, patient care is hampered by staff workload and subjective test interpretation. The use of automated systems is being widely explored for the removal of existing issues. Further, deep learning is emerging and giving promising results for quick and efficient image analysis. This paper gives a systematic review of automated multi-label classification of CXR images and a meta-analysis of the CheXpert CXR image dataset. This paper shows the comparison of the results of evaluation metrics like accuracy, F1-score, precision, and recall for multi-label classification by adopting different ways of handling the uncertainties for all 14 observations of the dataset. The performance of the model is assessed using test data that has not yet been observed. With no patients from the training set appearing in the test set, 200 trials from 200 patients were randomly selected from the whole dataset. Our experimental setup gives results that are an improvement upon earlier work; thus, this study will provide guidance as well as a higher accuracy level for the next chest radiography studies.