ESTRO 2021 Abstract Book

S1389

ESTRO 2021

PO-1666 Streamlining the use of PET/MRI in an MR-only radiotherapy workflow R. Winter 1 , N. Bach-Gansmo 1 , O. Sæther 2 , M. Alsaker 3 , K.R. Redalen 1

1 Norwegian University of Science and Technology, Department of Physics, Trondheim, Norway; 2 Trondheim University Hospital, Department of Radiology and Nuclear Medicine, St. Olavs hospital, Trondheim, Norway; 3 Trondheim University Hospital, The Cancer Clinic, St. Olavs hospital, Trondheim, Norway Purpose or Objective In radiotherapy an MR-only workflow is increasingly favored, as it provides in particular high-resolution images with better soft tissue contrast for delineation than CT. Hybrid PET/MR further extends the possibilities for tissue characterization delivering intrinsically co-registered data. However, to integrate PET/MR into the MR- only workflow two maps of the patient’s tissue are needed: a map of the linear attenuation coefficient m (mmap) for PET attenuation correction, and a map of CT Hounsfield units (HU) (pseudo CT; psCT) for dose calculation. Today, different MR sequences are used to generate the two maps. Our aim was to investigate if one MR sequence can be used for generating both maps to optimize the use of PET/MR in an MR-only workflow. Materials and Methods A 3T PET/MR scanner (Siemens) with radiotherapy specific positioning setup was used for pelvic and head/neck imaging of four volunteers (N=2, respectively). A Dixon MR attenuation correction (MRAC) sequence was used to generate a PET mmap (PETm_MRAC), and an anatomical T2-weighted sequence was used to generate a psCT using the MRI Planner software (psCT_MRIplanner) (Spectronic) (Fig.1-2 A&D). In addition, a PET mmap from the MRI Planner psCT (PETm_MRIplanner), and a psCT from the MRAC PET mmap (psCT_MRAC) were calculated using bilinear transformations between HU and m values (Fig.1-2 B&E). The purpose was to generate a PET mmap and a psCT from both MR sequences, i.e. to compare PETm_MRAC with PETm_MRIplanner and psCT_MRAC with psCT_MRIplanner. The mean absolute error (MAE) and structural similarity index measure (SSIM) were calculated within the body contour to assess agreement. Dice index compared different tissue classes defined in the mmaps. Results In pelvis, the MAE of volunteer #1/2 was 49.2/26.8 ·10 -4 cm -1 between PET mmaps, and 76.4/52.2 HU between psCTs. The mean SSIM was 0.988/0.994 between mmaps, and 0.995/0.996 between psCTs. The Dice index was 0.98/0.99 for the total anatomy, 0.62/0.83 for fat, 0.13/0.26 for the intermediate class between fat-water, 0.72/0.85 for water like soft tissue and 0.51/0.56 for bone. In head/neck, MAE was 35.8/57.8 ·10 -4 cm -1 between mmaps and 116.8/169.5 HU between psCTs. The mean SSIM was 0.984/0.977 between mmaps, and 0.989/0.987 between psCTs (Fig.1-2 C&F). The Dice index was 0.98/0.96 for the total anatomy, 0.81/0.74 for air-lung, 0.06/0.33 for fat, 0.34/0.03 for the intermediate class, 0.80/0.34 for water and 0.40/0.37 for bone. Conclusion When using hybrid PET/MR in an MR-only workflow the tasks of generating mmaps and psCTs are similar and have the potential for being streamlined. Even though our volunteer study showed that the procedures are very similar and one of the MR sequences may be sufficient, local differences exist in some tissue areas. Further work should investigate whether this will affect the radiation dose calculation or PET activity values. In this volunteer study, the psCTs could not be compared to real CTs; this is currently investigated in a clinical study.