RUOK: Recuration of a Public Dataset Utilized to Optimize Knowledge for Multi-label Chest X-ray Disease Screening

RUOK was implemented to aid the screening protocol at rural Thai hospitals, addressing a shortage of radiologists. Various public Chest X-ray (CXR) datasets focus on radiological abnormalities rather than diseases, limiting their suitability for training disease screening models. To address this limitation, we collaborated with expert radiologists to reclassify the labels of public datasets into eight classes representing prevalent diseases in Thailand: No finding, Suspected active tuberculosis, Suspected lung malignancy, Abnormal heart, great vessels, and mediastinum, Intrathoracic abnormal findings, Pneumonia, COVID-19, and Extrathoracic abnormal findings. This innovative adaptation of public dataset labels enhances the optimization of the data-efficient image transformers model for practical common diseases screening protocols. The Area Under the Receiver Operating Characteristics evaluated on test set yielded the results of 93.06%, 89.21%, 68.66%, 90.46%, 66.88%, 73.45%, 75.13%, and 80.63%, respectively. This highlights the model's potential in advancing disease classification and improving healthcare outcomes, particularly in regions with a shortage of radiological expertise.