55. A statistical method for assessing Low Contrast Detectability in CT: A phantom study using Filtered Back Projection and Iterative reconstructions

Purpose To calculate Low Contrast Detectability (LCD) curves using a statistical approach on CT images reconstructed with filtered back projection (FBP) and Iterative (IR) algorithms. Methods 25 axial images of a water phantom (25 cm diameter), scanned on a Somatom Definition Flash with different tube loads, were reconstructed with FBP and IR (IRIS) filters. A volume of 16 × 16 × 5 cm 3 (VOI-BIG) was divided in many smaller contiguous squared ROI (ROI-lets) drawn in each axial slice and the standard deviations (SDX-m) of their mean CT values (X-m) were calculated for different dimensions. Threshold (Th) for the detectability of a low contrast object can be predicted, for an ideal observer, as Th = 3.29 ∗ SDX-m, assuming a normal distribution of X-m with a confidence level of 95% and Signal-Known-Exactly/Background Known Exactly conditions. LCD curves were obtained with this method by increasing the dimensions of the ROI-lets from 2 mm (4 pixel) to 25 mm (52 pixel) for smooth, medium-smooth and sharp FBP filters together with their corresponding iterative versions at different dose levels (CTDIvol range: 2.8–20.7 mGy). Contrast Gain (CG) was defined as the difference between 2 LCD thresholds for the same object size. Results Compared to FBP images, the noise in IR with CTDIvol = 6.9 mGy, calculated as pixel SD in the VOI-BIG, was reduced by 24%, 33% and 36% for smooth, medium-smooth and sharp filter respectively. This improvement in image quality resulted also in the LCD curves: an average CG of 0.4 HU and 2.3 HU was observed for smooth/medium and sharp filters for details dimension within 5 and 20 mm. When FBP-Smooth and IR-Medium/Smooth filters are compared, CG increase monotonically with dose reduction and object size. Conclusions Although the proposed method is valid for ideal observers only, it provides a simple but potentially attractive metric for image quality assessment and protocol optimization.