Impact of resolution and noise characteristics of digital radiographic detectors on the detectability of lung nodules

Abstract

The current system performance metrics for Digital Radiographic detectors describe physical parameters, such as resolution (Modulation Transfer Function), noise (Noise Power Spectrum) and efficiency (Detective Quantum Efficiency). However, little has been done to substantiate the impact of these quantitative image quality metrics on a detector's utility for specific clinical tasks. In order to simulate the effects of these physical parameters, image modification routines were developed capable of modifying a perfect input image to the resolution and noise characteristics specified by an input MTF and input NPS and included sampling effects such as aliasing. Experimental verification of these routines showed excellent correspondence between the resolution and noise properties of the output images and the input NPS and MTF curves. In order to investigate the effect of noise and resolution on signal detection tasks, high-quality images containing simulated lesions are altered by the image modification routines to the resolution and noise properties of two commercial digital radiographic detectors, one direct and one indirect. The sets of modified images had noise properties consistent with acquisitions at comparable, clinically relevant exposures for the two detectors. An observer study is performed with the resultant images followed by a Receiver Operating Characteristic (ROC) analysis. The results revealed the direct detector had a higher area under the ROC curve with a statistically significant difference for a 2.75 mm nodule (A_z = 0.90 vs. 0.76, p<0.01). The findings illustrated the connection between the physical performance metrics and utility for the signal detection tasks necessary for clinical use.