Towards an atlas-based approach for including bone tissue in pelvic PET/MR attenuation correction

Abstract

Consulta en la Biblioteca ETSI Industriales (Riunet)

[EN] In hybrid Positron Emission Tomography (PET) imaging of prostate cancer, a trend to replace Computed Tomography (CT) by Magnetic Resonance Imaging (MRI) is gaining importance to improve the diagnostic quality. Unlike CT, however, MRI does not provide information on tissue density, which is required for correction of photon attenuation in PET. The most common approach for attenuation correction in PET/MRI is to use a segmented photon attenuation map, in which fixed attenuation coefficients are assigned to different tissue classes. However, bone tissue cannot be segmented on MR images and is therefore usually not taken into account. In the pelvic area the amount of bone is significant, which can lead to bias in the reconstruction PET images. The purpose of this study is to evaluate the effect of ignoring bone on Attenuation Correction (AC) and to start the development of an atlas-based method for MR-based attenuation correction considering bones in the pelvic area. Materials and Methods: In this project, 17 male patients were retrospectively analysed from lymphoma (12 patients) or lung cancer (5 patients) studies. These patients had been subjected to two consecutive examinations, one PET/CT scan and one PET/MR scan covering at least the pelvic area. Because of the focus on the pelvic area in study, we only analysed this part of the whole-body scans. The CT images were segmented into 3-class tissue (air, lung and fat), 4-class tissue (air, lung, fat and tissue), 5-class tissue (air, lung, fat, tissue and bone) and 6-class tissue (air, lung, fat, low-density bone and high-density bone) maps to investigate the effect of different segmentation methods (with and without taking into account bone) on AC. The reconstructed, attenuation corrected PET images were qualitatively and quantitatively analysed. For the latter, the bias in image intensity was evaluated in comparison with standard CT-based AC, which served as the gold standard.. Finally, based on these results, we analysed the possibility to develop an Atlas-based method for AC in PET/MR. Results: Qualitatively, no obvious differences were visually observed between the reconstructed PET images. Quantitatively, we found the mean ± standard deviation (root mean squared error) bias to be 7.6 ± 7.6% (10.8%) for 3C, 0.8 ± 3.2% (3.4%) for 4C, -0.5 ± 2.3% (2.6%) for 5C and -0.5 ± 1.9% (2.3%) for 6C in fat, -1.5 ± 5.5% (7.1%) for 3C, 2.5 ± 4.2% (4.9%) for 4C, -0.9 ± 2.2% (2.6%) for 5C and -1.0 ± 1.4% (1.9%) for 6C in tissue, and -18.7 ± 13.3% (23.7%) for 3C, -14.3 ± 13.4% (19.8%) for 4C, -3.1 ± 11.7% (12.4%) for 5C and -2.3 ± 5.7% (6.3%) for 6C in bone. Conclusion: The 3C and 4C models used currently in PET/MRI AC result in large negative bias of the reconstructed activity in bones. The 6C model is the best for PET reconstruction as it shows only small differences with gold standard. For these reasons, the creation of an Atlas-based method taking into account pelvic bone has the potential to substantially improve AC in pelvic PET/MRI.