MENU

Click here to

×

Are you sure ?

Yes, do it No, cancel

Whole-Body PET Estimation From Low Count Statistics Using Deep Convolutional Neural Networks

X Dong1, Y Lei1 , T Wang1 , K Higgins1,2 , T Liu1,2 , W Curran1,2 , H Mao2,3 , J Nye3* , X Yang1,2 , (1) Department of Radiation Oncology, Emory University, Atlanta, GA (2)Winship Cancer Institute, Emory University, Atlanta, GA (3)Department of Radiology and Imaging Sciences, Emory University, Atlanta, GA

Presentations

(Sunday, 7/14/2019) 4:00 PM - 6:00 PM

Room: 304ABC

Purpose: Lowering either the administered activity or scan time of PET imaging is desirable as it decreases the patient’s radiation burden or can improve patient comfort and reduce motion artifacts. But reducing these parameters lowers overall photon counts and increases noise, adversely impacting image contrast and quantification. To address this low count statistics problem, we propose a novel 3D cycle-consistent generative adversarial network (GAN) model to estimate diagnostic quality PET images using low count data.

Methods: Cycle GANs learn a transformation to synthesize diagnostic PET data using low count PET data that would be indistinguishable from our standard clinical protocol. The algorithm also learns an inverse transformation such that the cycle low count PET data (inverse of the synthetic estimate) generated from synthetic full count PET is close to the real low count PET. To optimize the matching between the synthetic and their cycle to their respective real datasets, both transformations are implemented by a generator network and their outputs are judged by a discriminator. We introduced residual blocks into the generator to catch the differences between low count and full count PET in the training dataset. Twenty-five subjects with whole-body PET/CT were retrospectively processed to derive diagnostic quality PET datasets from 1/8 count of the standard PET protocol. These data were then compared to the original diagnostic PET.

Results: The average ME and NMSE in whole body were -0.14%±1.43% and 0.52%±0.19% with the proposed cycle consistent GAN model, comparing to 5.59%±2.11 and 3.51%±4.14% on the original low count PET images. NCC is improved from 0.970 to 0.996, and PSNR is increased from 39.4 dB to 46.0 dB.

Conclusion: We developed a novel learning-based approach to accurately estimate diagnostic quality PET datasets from one eighth of the photon counts based on a 3D cycle-consistent GAN with integrated residual blocks.

Keywords

Not Applicable / None Entered.

Taxonomy

Not Applicable / None Entered.

Contact Email