Development of advanced denoising and analysis algorithms for applications in hybrid PET/MRI brain imaging

Hybrid PET/MRI scanners are becoming more common in research and clinical settings, in particular for their ability to simultaneously acquire unique functional and structural information to probe the healthy and diseased brain. Data from both modalities need to be thoroughly processed to enhance signal-to-noise ratios (SNR), and the development and optimization of analysis algorithms is required to extract meaningful physiological quantities for medical research applications. This work provides significant contributions to these fields through the development, testing, and validation of four algorithms. PET data are intrinsically noisy due to the Poisson nature of positron emission events and the relatively low fraction of events detected by scanners, so we first improve upon an existing PET denoising algorithm to overcome limitations in quantitative accuracy in precision.

This proves important in task-based PET imaging, where tasks presented to a subject during scanning result in a small temporal signal change that is difficult to distinguish from noise. The developed PET denoising algorithm is thus coupled with an advanced PET detection algorithm which provides a significant improvement in the detection sensitivity of task-induced signal changes. Motivated by the success of existing PET denoising algorithms, we then develop an MRI denoising algorithm for low SNR functional MRI data, which demonstrates significant improvements in quantitative accuracy and precision of downstream modeling metrics.

Finally, we propose a pattern analysis extension of an existing joint PET/MRI analysis algorithm to identify disease-related alterations to simultaneously acquired multimodal data. We identify immediate applications of each algorithm in novel medical research studies, and proof-of-concept testing on human data is performed when available.