Premium
Investigating the use of nonattenuation corrected PET images for the attenuation correction of PET data
Author(s) -
Chang Tingting,
Diab Rami H.,
Clark John W.,
Mawlawi Osama R.
Publication year - 2013
Publication title -
medical physics
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.473
H-Index - 180
eISSN - 2473-4209
pISSN - 0094-2405
DOI - 10.1118/1.4816304
Subject(s) - attenuation , voxel , correction for attenuation , thresholding , artificial intelligence , computer science , segmentation , nuclear medicine , computer vision , physics , optics , medicine , image (mathematics)
Purpose: The aim of this study is to investigate the feasibility of using the nonattenuated PET images (PET‐NAC) as a means for the AC of PET data.Methods: A three‐step iterative segmentation process is proposed. In step 1, a patient's body contour is segmented from the PET‐NAC using an active contour algorithm. Voxels inside the contour are then assigned a value of 0.096 cm −1 to represent the attenuation coefficient of soft tissue at 511 keV. This segmented attenuation map is then used to correct for attenuation the raw PET data and the resulting PET images are used as the input to Step 2 of the process. In step 2, the lung region is segmented using an optimal thresholding approach and the corresponding voxels are assigned a value of 0.024 cm −1 representing the attenuation coefficients of lung tissue at 511 keV. The updated attenuation map is then used for a second time to correct for attenuation the raw PET data, and the resulting PET images are used as the input to step 3. The purpose of Step 3 is to delineate parts of the heart and liver in the lung contour using a region growing approach since these parts were unavoidably excluded in the lung contour in step 2. These parts are then corrected by using a value of 0.096 cm −1 in the attenuation map. Finally the attenuation coefficients of the bed are included based on CT images to eliminate the impact of the couch on the accuracy of AC. The final attenuation map is then used to AC the raw PET data and generates the final PET image, which we name iterative AC PET (PET‐IAC). To assess the proposed segmentation approach, a phantom and 14 patients (with a total of 55 lesions including bone) were scanned on a GE Discovery‐RX PET/CT scanner. PET‐IAC images were generated using the proposed process and compared to those of CT‐AC PET (PET‐CTAC). Visual inspection, lesion SUV, and voxel by voxel histograms between PET‐IAC and PET‐CTAC for phantom and patient studies were performed to assess the accuracy of image quantification.Results: Visual inspection showed a small difference in lung parenchyma between the PET‐IAC and PET‐CTAC. Tumor SUV based on PET‐IAC were on average different by 3% ± 9% (6% ± 7%) compared to the SUVs from the PET‐CTAC in the phantom (patient) studies. For bone lesions only, the average difference was 3% ± 6%. The histogram comparing PET‐CTAC and PET‐IAC resulted in an average regression line of y = (1.08 ± 0.07) x + (0.00007 ± 0.0013), with R 2 = 0.978 ± 0.0057.Conclusions: Preliminary results suggest that PET‐NAC for the AC of PET images is feasible. Such an approach can potentially be used for dedicated PET or PET/MR hybrid systems while minimizing scan time or potential image artifacts, respectively.