PROJECT TITLE :

mDixon-Based Synthetic CT Generation for PET Attenuation Correction on Abdomen and Pelvis Jointly Using Transfer Fuzzy Clustering and Active Learning- Based Classification

ABSTRACT:

To generate synthetic CT images from modified Dixon (mDixon) MR data, we have devised a new approach. When reconstructing PET data from the abdomen and pelvis, the synthetic CT is employed to adjust for attenuation (AC). AC is essential in PET/MR systems in order to be quantitatively accurate and meet qualification criteria required for use in many multi-center trials, even though it isn't a part of MR itself. Anatomical variety in patients with diseases is difficult to account for in existing MR-based synthetic CT approaches, which either rely on long acquisition times and restricted clinical availability, or on the matching of newly-scanned subjects to previously-scanned subjects in an MR-CT library. It's a five-phase interlinked strategy that leverages powerful machine learning techniques for synthetic CT generation to address these shortcomings. A combination of transfer fuzzy clustering and TFC-ALC is employed. Our initiatives have a four-fold significance: It is possible to generate a superior synthetic CT image with TFC-ALC than with Dixon-based scanning, which is already in use on the difficult abdomen. TFC uses transfer learning to divide MR voxels into four groups: fat, bone, air, and soft tissue. ALC can develop insightful classifiers, employing as few but informative labelled examples as feasible, to accurately discriminate bone, air, and soft tissue. The TFC-ALC method successfully overcomes the inherent imperfection and potential uncertainty about the co-registration between CT and MR images by combining the two techniques. 3) TFC-improved ALC's parameter robustness and preferred synthetic CT production make it more clinically applicable than existing approaches. Using mDixon-MR data from ten participants, the proposed technique resulted in an average score of mean absolute pre...