Online image corrections applied to a dedicated breast PET. (16th March 2016)

• Record Type: Journal Article
• Authors: Moliner, L.; González, A.J.; Correcher, C.; Benlloch, J.M.
• Is Part Of: Journal of instrumentation. Volume 11:Number 3(2016:Mar.)
• Page Start: C03035
• Page End: C03035
• Publication Date: 2016-03-16
• Subjects: Image reconstruction in medical imaging -- Medical-image reconstruction methods and algorithms, computer-aided software; Scientific apparatus and instruments -- Periodicals; 502.84
• DOI: 10.1088/1748-0221/11/03/C03035
• Languages: English
• ISSNs: 1748-0221

Abstract

Abstract: In this work, we present the online implementation of attenuation, scatter and random corrections using the LMEM algorithm for the dedicated breast PET named MAMMI. The attenuation correction is based on image segmentation, the random correction is derived from the rate estimation of single photon events and the scatter correction is determined by the dual energy window method. These three corrections are estimated and implemented in the reconstruction process without almost increasing the reconstruction time. The image quality is evaluated in terms of image uniformity and contrast using the reconstructed images of two custom-designed phantoms. When we apply the three corrections, the measured uniformity in the whole field of view is (10 ± 1)% compared to (17 ± 1)% without corrections. The adapted recovery contrast coefficients (normalized to 1) are approximately (0.80 ± 0.02) in hot areas, improving the value of (0.66 ± 0.07) obtained without corrections. The reconstruction processing time is also studied, finding an increment of around 7% when the three corrections are simultaneously included. Finally, 25 breast image datasets are also analyzed. The average acquisition time per patient is around 1200 seconds and the reconstruction times with corrections vary from 100 to 400 seconds using (1 × 1 × 1) mm 3 voxel size and from 300 to 1800 seconds using (0.5 × 0.5 × 0.5) mm 3 voxel size. These reconstructions are performed with a virtual pixel size of (1.6 × 1.6) mm 2 and twelve iterations.