Hybrid PET/MRI systems are very advantageous for a variety of clinical applications by combining the soft tissue contrast of MRI with the functional and metabolic information of PET. These systems have found success for oncology studies, particularly in head and abdomen/pelvis, as well as for epilepsy, neurological diseases, heart disease, and pediatrics for dose reduction. However, the PET resolution and SNR is typically worse than MRI, and suffers from the loss of feature and data due to motion as well. PET/MRI systems offer the potential to create more accurate, higher resolution PET reconstructions, including correction of artifacts, motion, and im- proved localization, by performing synergistic reconstructions that leverage the simultaneous data acquisition. In particular, this fellowship proposes to develop novel physics-constrained machine learning models for informa- tion sharing between PET and MRI for enhanced spatial localization, estimation of attenuation and activity, and motion. We propose to develop a deep maximum-likelihood estimation of attenuation and activity (MLAA) that can compensate for artifacts and improve PET reconstruction accuracy. We also propose a motion-enhanced joint PET/MRI reconstruction to capture arbitrary motions and reduce dose requirements for chest and abdomen studies. Together, these models aim to improve the PET spatio-temporal resolution, SNR, and quanti?cation for a broad range of clinical applications, and will be evaluated for cancer assessment in the pelvis, liver, and lung. This fellowship will be performed in the Department of Radiology and Biomedical Imaging at UCSF under the guidance of Prof. Peder Larson, who leads a research program on advanced imaging methods development, and Dr. Thomas Hope, a radiologist and nuclear medicine physician who leads multiple PET/MRI projects. The Department is one of the leading centers in biomedical imaging research, and has been at the forefront on translating PET/MRI systems into clinical practice. The UCSF PET/MRI scanner has dedicated research time, which is also available on other MRI and PET/CT research systems, and extensive computational resources to support the proposed project. The applicant, Dr. Abhejit Rajagopal, has a background in computational imaging and machine learning, will be jointly mentored by this engineer/physician team. He will be trained to become a biomedical imaging scientist by participating in formal coursework on medical imaging systems, training on the PET/MRI system, grant writing, and performing clinical research, supporting his development into a creative, independent biomedical researcher.
Hybrid positron emission tomography (PET) and magnetic resonance imagery (MRI) imaging systems cur- rently aid in diagnosis and prognosis of numerous types of cancer and disease, but are not always precise enough to accurately measure and track a patient?s response to therapy, particularly in organs and tissue that are sub- ject to motion. There is an unrealized potential here to synergistically combine complimentary PET-MRI data to dramatically improve the spatial resolution and SNR of PET, as well as to create motion-resolved 4D (x,y,z,t) imagery by combining information across modalities and time-frames to combat severe undersampling. These methods will be evaluated in human studies of cancer to capture ?ne structure and micro-features on moving organs (e.g. lung nodules, liver metastases), ultimately aiding in quantitative characterization of disease.
