Abstract
371
Objectives The authors proposed to assess lesion detection on respiratory-gated clinical PET/CT using a novel 4D numerical observer, and to compare the lesion detectability using gated, motion-corrected, and non-gated methods.
Methods The 4D observer was developed with two-staged processing: 1) accommodating the spatial domain in each gated image using a set of 3D Channelized Hotelling Observers (CHO), and 2) integrating all the spatial and temporal information using a Hotelling Observer. Each gate was locally prewhitening and matched filtering to decorrelate and detect the presence of spatial signals. Then, the intermediate scores computed from all the CHOs were integrated to finally assess lesion detectability accounted both changes in lesion location and shape, which mimicked gated cine viewing. Six patients were acquired with respiratory gating for both gated 18F-FDG PET and gated CT. The hybrid images of ‘lesion-present’ were generated, which were sufficient realism and able to characterize the presence and location of each lesion. Four 13-mm spherical lesions were simulated and added to the clinical data. The ROIs of lesion-present and -absent OSEM images were computed detection SNR, and compared for:1) the 4D observer for 8-binning gated (4D-N-G), 2) 3D-CHO for image-based motion-corrected (3D-CH-MC), and 3) 3D-CHO for non-gated (3D-CH-NG) methods.
Results On average, 4D-N-G improved the detection SNR by 491.9% (p<0.01), whereas 3D-CH-MC improved by 217.7% (p<0.07) as compared to 3D-CH-NG. The relative SNR (Gain=SNRx/SNR3D-CH-NG) of 4D-N-G was significantly higher than that of 3D-CH-MC by 160.0% (p<0.03), where Gain4D-N-G was 4.92±1.96 and Gain3D-CH-MC was 2.18±1.57.
Conclusions The proposed 4D observer provides significant improvement in lesion detectability on respiratory-gated clinical PET over the 3D approaches of motion-corrected and non-gated images.
Research Support R01HL110241,R01CA165221,R01EB013293.