Search

Magnetic resonance imaging (MRI) guided attenuation correction (AC) of positron emission tomography (PET) data and/or radiation therapy (RT) treatment planning is challenged by the lack of a direct link between MRI voxel intensities and electron density. Therefore even if this is not a trivial task a pseudo computed tomography (CT) image must be predicted from MRI alone. In this work we propose a two step (segmentation and fusion) atlas based algorithm focusing on bone tissue identification to create a pseudo CT image from conventional MRI sequences and evaluate its performance against the conventional MRI segmentation technique and a recently proposed multi atlas approach. The clinical studies consisted of pelvic CT PET and MRI scans of 12 patients with loco regionally advanced rectal disease. In the first step bone segmentation of the target image is optimized through local weighted atlas voting. The obtained bone map is then used to assess the quality of deformed atlases to perform voxel wise weighted atlas fusion. To evaluate the performance of the method a leave one out cross validation (LOOCV) scheme was devised to find optimal parameters for the model. Geometric evaluation of the produced pseudo CT images and quantitative analysis of the accuracy of PET AC were performed. Moreover a dosimetric evaluation of volumetric modulated arc therapy photon treatment plans calculated using the different pseudo CT images was carried out and compared to those produced using CT images serving as references. The pseudo CT images produced using the proposed method exhibit bone identification accuracy of 0.89 based on the Dice similarity metric compared to 0.75 achieved by the other atlas based method. The superior bone extraction resulted in a mean standard uptake value bias of??-1.5??±??5.0 (mean??±??SD) in bony structures compared to??-19.9??±??11.8 and??-8.1??±??8.2 achieved by MRI segmentation based (water only) and atlas guided AC. Dosimetric evaluation using dose volume histograms and the average difference between minimum/maximum absorbed doses revealed a mean error of less than 1 for the both target volumes and organs at risk. Two dimensional (2D) gamma analysis of the isocenter dose distributions at 1/1?mm criterion revealed pass rates of 91.40??±??7.56 96.00??±??4.11 and 97.67??±??3.6 for MRI segmentation atlas guided and the proposed methods respectively. The proposed method generates accurate pseudo CT images from conventional Dixon MRI sequences with improved bone extraction accuracy. The approach is promising for potential use in PET AC and MRI only or hybrid PET/MRI guided RT treatment planning.