The performance of digital x-ray imaging systems in detection of subtle lung nodules.

This dissertation addresses the diagnostic effectiveness of chest radiography with regard to the detection of subtle lung nodules (SLNs). First, the contrast characteristics of SLNs were determined from radiographs containing confirmed lesions. The nodules were found to have poorly defined edges and diverse variability in contrast (0.02-0.13) and diameter (3-16 mm). A functional form of low spatial frequency content was found to adequately fit the subject contrast profiles of the SLNs. These radiologic properties were then used to develop a set of Teflon nodule phantoms with similar x-ray transmission and variability, to be utilized in human observer experiments. Secondly, the physical performance of a digital chest radiographic system (Fuji FCR-9501/FCR-9501-HQ) was determined by assessing its resolution (modulation transfer function--MTF), noise (noise power spectrum--NPS), and signal-to-noise (detective quantum efficiency--DQE) characteristics. The MTF was measured using an edge method, the accuracy of which was independently verified against an established slit method. The NPS measurements were performed by 2D Fourier analysis of images acquired with uniform exposures. The DQE(f) of the system was determined from these measures along with exposure data and computed ideal signal-to-noise ratios. Measurements were performed at three kVps, 70, 115, and 150. The results indicate that the resolution response of the imaging system minimally blurs the appearance of SLNs, while the NPS and DQE are such that for a typical chest exposure, even the smallest SLNs should be visible. Finally, the effect of the background lung structure on the detection of SLNs was explored. Simulated nodules with contrast characteristics determined earlier in the study were superimposed centrally on images extracted from digital chest radiographs, and from digital radiographs containing only quantum mottle associated with typical chest exposures. Two observer performance studies were undertaken. For lung tissue backgrounds, the detection threshold ($A\sb{z} = 0.8$) was achieved for tissue-equivalent nodules of 4.5 mm (4.4% peak contrast). A corresponding value of 1 mm (1.1% peak contrast) was determined for quantum noise backgrounds. The results suggest that anatomical noise associated with the background lung structure is the main factor limiting the detection of SLNs in chest radiographs.Ph.D.Applied SciencesHealth and Environmental SciencesMedical imagingNuclear engineeringUniversity of Michigan, Horace H. Rackham School of Graduate Studieshttp://deepblue.lib.umich.edu/bitstream/2027.42/130350/2/9722081.pd