TU-A-12A-05: Evaluating the Consistency of Lesion Volumetry in Chest CT After Substantial Dose Reduction




To quantify how the choice of dose affects semi-automated lung lesion volume measurements on chest CT.


We created a framework consisting of three parts: (1) software to simulate reduced-dose sinograms from real clinical CT scans, (2) semi-automated contouring tools for measuring lesion volumes, and (3) figures of merit to quantify reader performance. Three readers measured 35 lung lesions on images acquired with our routine chest protocol (Siemens Definition Flash, 120 kV, 250 reference mAs, B45f kernel) and four simulated fractions of the clinical protocol (25%, 10%, 3%, and 1%). To quantify reader performance, we compared the percent differences between clinical and reduced-dose volumes against 15% and 30% thresholds, inspired by a QIBA claim about the reliability of tumor volume change measurements. We analyzed the distribution of percent differences at each dose, as well as outliers in those distributions.


The 35 clinical lesions produced volumes ranging from 150 mm^3 to 22500 mm^3, corresponding to effective spherical diameters of 7 to 35 mm. The distributions of percent differences appeared consistent across the dose levels we considered, with the majority of absolute percent differences less than 15%. Some exceptions to this rule were smaller lesions that did not comply with QIBA’s definition of a clinically-relevant lesion.


Our framework shows promise for quantifying the effects of dose on various clinical tasks. Based on the relatively small patient cohort in this study, the task of measuring clinicallyrelevant lesion volumes seemed to be independent of dose down to the 1% level (corresponding to a phantom CTDIvol value of 0.21 mGy), with the exception of a few outliers that did not comply with the QIBA definition of measurable, clinically-relevant lesions. For QIBA-compliant lesions, the dose level does not appear to affect reader performance.