Abstract W P28: Precision of Tmax and CBF Lesion Volume Estimates in CT Perfusion Imaging
Objective: CT perfusion (CTP) imaging is increasingly used to select patients for acute stroke trials. Patient selection criteria are typically based on the volumes of reversibly and irreversibly injured tissue. Modern software allows for automated and quantitative assessment of these lesion volumes. There is a push towards reducing the radiation dose used in CTP protocols which may adversely affect the reproducibility of lesion volume assessments. The aim of this study was to assess the effect of radiation dose dependent image noise on the precision of lesion volume estimates.
Methods: We generated 6 synthetic noise-less datasets based on a brain atlas of gray matter, white matter and CSF and integrated co-registered perfusion lesions from 6 acute stroke patients. For each pixel, concentration time curves were simulated using standard tracer kinetic models to obtain CTP source data based on the lesions of the 6 patients. To simulate a CTP protocol with standard radiation dose, we added CT textured noise (“Standard noise”) with a SD of 10 Hounsfield Units (HU) to the dynamic datasets and to simulate a low-dose (“high noise”) CTP protocol we added noise with a SD of 14 HU. This step was repeated 50 times per noise level and for each iteration we determined, using automated perfusion software, the volumes of the ischemic core (rCBF based) and of critically hypoperfused tissue (Tmax>6s).
Results: With standard noise, mean CBF volume was 5.6 ml (SD 2.1) and mean Tmax>6s volume was 38.2 ml (SD 1.3). In the high noise condition, the Tmax>6s lesion volume was on average 13% (5 ml) larger (p<0.0001) than in the standard noise condition. The SD of the Tmax volumes increased from 1.3 mL to 1.8 mL with higher noise. Conversely, with high noise there was an average 5 mL (range 0-16 mL) absolute reduction in CBF lesion volume as compared to the low noise condition (p<0.0001). The CBF lesion became in many cases undetectable (estimated as 0 ml) in the high noise setting.
Conclusions: Noise in CTP data influences the precision of lesion volume estimates. The estimate of the infarcted core seems particularly sensitive to noise. This knowledge should be used to develop guidelines that specify the maximum amount of noise (ie minimum amount of radiation) that is acceptable in a given CTP protocol.
Author Disclosures: S. Christensen: Consultant/Advisory Board; Significant; Consulting for IschemaView Inc.. N. Mishra: None. M. Straka: None. D. Fleischmann: None. R. Bammer: Ownership Interest; Significant; IschemaView Inc. G. Zaharchuck: None. M. Parsons: None. C. Levi: None. G. Albers: Ownership Interest; Significant; IschemaView Inc. M. Landsberg: None.
- © 2014 by American Heart Association, Inc.