Author: Matthew R. Hoerner, Allison Shields 👨🔬
Affiliation: Yale University School of Medicine, Yale University 🌍
Purpose: To investigate the quality and clinical utility of chest x-rays synthesized from CT scans (sCXR).
Methods: Five helical chest CT exams were chosen for evaluation: this cohort represented a subset of ICU patients who had companion x-rays within 24 hours of the CT acquisition, where 4/5 were acquired on a portable unit. The CT table was automatically segmented from the images, and x-ray projections were generated in the posterior-anterior and left-lateral orientations. The images were fed into a multi-frequency processing framework, following that of conventional radiography. For-presentation images were uploaded to a test server, along with the companion chest radiograph(s), CT localizers, and axial CT study. Images were then qualitatively assessed by three radiologists using a 3-point Likert scale, focusing on the lungs, hilum/mediastinum, spine/ribs, diaphragm, and tubes/lines/devices. The radiologists also evaluated the presence of any artifacts, and whether a lateral sCXR provided additional information for diagnosis. 2D-MTF measurements were taken using the slanted edge method as a quantitative assessment of spatial resolution between the localizer and sCXR.
Results: Radiologist reviewers found the sCXR’s equivalent if not superior to the chest radiograph and localizer in most cases, with superior visualization of the diaphragm, spine, and ribs. Some breathing motion is visible in the simulated images, which contributed to blurring artifacts in the tubes/lines/devices category. Reviewers agreed that the lateral sCXR provided additional information for diagnosis. The 2D-MTF comparison indicated superior spatial resolution along the X-axis compared to the localizer, but roughly equivalent resolution along the slice direction with the current projection geometry. Z-axis resolution was lower in both instances due to table motion.
Conclusion: This work illustrates the clinical utility of generating simulated chest radiographs from CT’s. Future investigations will utilize a greater number of scans across a wider range of clinical indications, in addition to application-specific processing techniques.