Keywords:
Artificial Intelligence, Radiographers, Thorax, Digital radiography, Education, Perception image, Technology assessment, Education and training, Quality assurance
Authors:
Y. Meng, J. Ruan, X. Y. Gong
Purpose or learning objective
Chest radiography is the most commonly performed diagnostic radiographic examination [1]. Image quality assessment plays a vital role in improving image quality and diagnostic accuracy[2]. Quality assessment (QA) for chest radiographs is still a manual process, which leads to poor interobserver agreement and limits the number of images included in the evaluation[3], further hindering the analysis of the overall image quality in the health care institution.
Among the factors affecting the quality of chest radiography, layout and position errors are common image acquisition defects that cannot be adjusted by postprocessing[4]. In recent years, deep learning algorithms have been widely used for lung nodule detection, automatic measurement of bones and joints, and image segmentation in radiology research [5-7], bringing substantial improvements to computer-aided diagnosis. Deep learning for image quality control is mainly limited to optimizing image quality[8] and lacks application for image quality assessment.
To our knowledge, few studies have quantified the image layout and position of chest radiographs. The aim of this study was to construct objective and quantitative assessment indices for the image layout and position of chest radiographs. Subsequently, we developed a fully automatic system for assessing image quality based on deep learning and linear regression cascade algorithms, as well as validating the performance of the system.