Keywords:
Computer applications, MR physics, Paediatric, CT, Image manipulation / Reconstruction, MR, Computer Applications-3D, Radiation safety, Dosimetric comparison, Image verification
Authors:
M. Alhilani1, H. Jeong2, G. Ntolkeras2, S. R. Atefi2, L. Zöllei2, A. Pourvaziri2, M. H. Lev2, E. P. Grant2, G. Bonmassar2; 1London/GB, 2Boston, MA/US
DOI:
10.26044/ecr2021/C-10139
Methods and materials
Development and validation of the numerical model:
- The first step was to select a patient with available whole-body sequences without major deformities as well as multiple sequences that would facilitate the segmentation process. Whole-body MRIs were acquired using MRRAGE, T2 FLAIR, Inversion Recovery, T1, Echo Planar Fast Spin Echo, T2 Fast Spin Echo, Radial VIBE sequences, and dual-energy CT scans, which were used to segment the various tissue compartments. We then pulled all the available MRI and CT scans from the database of the Boston Children’s Hospital.
- The meticulous process of the segmentation of the different tissues of the body was slightly differentiated for the brain and the non-brain tissues. (a) non-brain tissues were segmented manually with 3D Slicer,7 using different sequences by two medical doctors.
(b) For the segmentation of the brain structures, an automated infant-specific data processing framework was used,6 and the results were reviewed by two specialized neuroradiologists.
- When the agreement was achieved between the operators and the tissues were aligned with the surrounded segmented tissues, the two sub-speciality certified neuroradiologists confirmed the result following a “pass or fail” process through an interactive process. The tissues that were scored with a “pass” were then finalized. All the tissues were put together in the reference T1 sequence, and the model was again confirmed by the two neuroradiologists.
- Furthermore, the MARTIN model itself was validated using three separate approaches. First, the calculated volume, weight, and dimensions of each structure were all within 6% of the literature values in the 29-month-old age group. Second, the results of the segmentation were also validated to minimize the error and bias by two subspecialized neuro-radiologists (step 3). Final validation was performed through a Bloch simulator to compare the segmentation quality of the synthetic MRI contrast of the MARTIN model to the original MRI data, which resulted in a structural resemblance of 0.9691.