Validation of Algorithmic CT Image Quality Metrics with Preferences of Radiologists

Abstract

Automated assessment of perceptual image quality on clinical Computed Tomography (CT) data by computer algorithms has the potential to greatly facilitate data-driven monitoring and optimization of CT image acquisition protocols. Its application in clinical operation necessitates the investigation of the correspondence of the algorithmic measurements to clinical expectations. This study addressed the need to validate algorithmic image quality measurements on clinical CT images with preferences of radiologists and determine the clinically acceptable range of algorithmic measurements.Algorithmic measurements of image quality metrics (organ HU, noise magnitude, spatial resolution, and noise texture) were performed on a clinical CT image dataset using techniques investigated previously. Algorithmic metric of clarity was defined and measured. The algorithmic measurements of noise magnitude, liver parenchyma HU, and clarity were compared to clinical expectations of image quality by radiologists in an observer study. Sets of CT liver images were selected from the dataset where images in the same set varied in terms of one metric at a time. These sets of images were shown via a web interface to a radiologist (observer) one set at a time. The observer first rank ordered the CT images in a set according to their preference for the varying metric. The observer then selected his/her preferred acceptable range of the metric within the ranked images. The rank order data were analyzed to investigate the agreement between algorithmic and observer assessments. The acceptable range data in observer rankings were analyzed to determine the clinically acceptable thresholds of algorithmic metric values.The overall rank order agreements between algorithmic and observer assessments were (perfect agreement has value of 1): 0.90, 0.98, and 1.00 for noise magnitude, liver parenchyma HU, and clarity respectively. The results indicate a strong agreement between algorithmic and observer assessments of image quality. Clinically acceptable thresholds (median) of algorithmic metric values were: (17.8, 32.6) HU for noise magnitude, (92.1, 131.9) for liver parenchyma HU and (0.45, 0.55) for clarity. The correspondence of algorithmic image quality assessments with clinical expectations paved the way towards establishing diagnostic reference levels in terms of clinically acceptable perceptual image quality and data-driven monitoring and optimization of CT image acquisition protocols.