Optimization of Tau PET Imaging for Alzheimer's Disease through Deep Learning-Based Image Reconstruction

点击翻译按钮获取中文摘要

Abstract Alzheimer’s disease (AD) is a progressive neurodegenerative disease characterized by memory loss, cognitive impairments, and behavioral disorders. 6.2 million people aged 65 and older are living with AD in the United States in 2021. Earlier diagnosis of AD holds particular significance as therapies are most effective during the pre-symptomatic stages before irreversible brain damage has occurred. Tau neurofibrillary tangles (NFTs), accumulating decades before symptomatic onset, can indicate the pre-symptomatic stages. According to Braak staging, tau NFTs start from transentorhinal, then spreading to hippocampus and other cortices at later stages. Detecting tau NFTs during early stages and clearly resolving their patterns is essential for early diagnosis and treatment monitoring of AD. With recent breakthroughs in tau tracer developments, Positron Emission Tomography (PET) can detect accumulation of tau NFTs in vivo. However, due to signal-to-noise ratio (SNR) and resolution limits of PET, accurate recovery of tau retention patterns in thin cortical regions is difficult. This is especially true for early stages when tau signal is weak. Additionally, recent longitudinal studies show that the accumulation change of tau deposits detected by PET is around 3 to 6 % per year for the AD group, and less for the preclinical AD group. This small annual change further challenges the signal detectability of current PET systems. Furthermore, 18F-MK-6240 is a newly developed tau tracer with higher affinity to tau NFTs and no off- target bindings near early Braak-staging regions, which makes it highly promising for early AD diagnosis. However, one issue with 18F-MK-6240 is the off-target bindings in the meninges. Given the thin nature of the cortical ribbon and its proximity to the meninges, quantitative accuracy of tau accumulation is significantly compromised. Consequently, there are unmet needs to further improve PET resolution and SNR for tau imaging. This grant application proposes deep learning (DL)-based image reconstruction methods that can improve the resolution and signal-to-noise ratio (SNR) of tau imaging. The four specific aims of this proposal are (1) to develop DL-based static PET image reconstruction; (2) to develop DL-based image reconstruction for dynamic PET; (3) to develop frameworks that can rapidly produce high-quality parametric images; and (4) to apply the proposed frameworks to 18F-MK-6240 imaging datasets. We expect the integrated outcome of the specific aims will be robust and clinically effective frameworks that can generate static and parametric images with improved resolution and SNR from static and simplified dynamic tau PET imaging.