MENU

Click here to

×

Are you sure ?

Yes, do it No, cancel

On the Quantification of PET Images for Treatment of Lung Cancer Patients

H Zhong*, N Morrow, J Kim, Medical College of Wisconsin, Rockford, IL

Presentations

(Sunday, 7/12/2020)   [Eastern Time (GMT-4)]

Room: AAPM ePoster Library

Purpose: A motion model-enhanced PET image reconstruction method may reduce motion artifacts without increasing data acquisition time, but standardized uptake values (SUVs) in resultant images could be compromised by errors in the motion model. The purpose of this study is to estimate the impact of the modeling errors on reconstructed PET images.

Methods: A blurry image decomposition (BID) algorithm was developed that uses a set of deformation maps to decompose a blurry image into a set of motion-freeze images. To evaluate the BID algorithm, a set of 10-phase 4D-PET and 10-phase 4DCT images were acquired sequentially from a lung cancer patient. A B-Spline-based deformable image registration (DIR) algorithm and hybrid DIR algorithm were applied to the 4DCT images to generate deformation maps. The 4D-PETs were down-sampled by merging every two adjacent transversal slices to create low-resolution PETs. The BID algorithm was applied to the original and resampled average PETs, and the quality of the reconstructed images was evaluated using signal-to-noise ratio (SNR) and universal quality index (UQI), respectively.

Results: The maximum SUVs and gross tumor volumes measured from all 4D-PETs were 20.6±1.14 and 250.2±5.6 cm3, and the maximum SUVs in the down-sampled 4D-PETs are 19.9±1.03 on average. The mean SUV differences between the 4D-PET at phase 0 and the reconstructed PETs at the high and low resolutions were 0.13 and 0.11 for the BSpline-based DIR maps, and 0.12 and 0.10 for the hybrid DIR maps. The down-sampling operations improved the quality of the reconstructed PETs with UQI increased from 0.91 to 0.94, and SNR increased by 17.5 % on average. An analytical comparison also showed that registration errors less than half a PET voxel have no impact on the BID algorithm.

Conclusion: The model-enhanced BID algorithm combined an image resampling strategy may help improve the quantitative accuracy of PET images.

Keywords

Not Applicable / None Entered.

Taxonomy

Not Applicable / None Entered.

Contact Email