Improved amyloid burden quantification with nonspecific estimates using deep learning

Abstract

Purpose Standardized uptake value ratio (SUVr) used to quantify amyloid-β burden from amyloid-PET scans can be biased by variations in the tracer’s nonspecific (NS) binding caused by the presence of cerebrovascular disease (CeVD). In this work, we propose a novel amyloid-PET quantification approach that harnesses the intermodal image translation capability of convolutional networks to remove this undesirable source of variability. Methods Paired MR and PET images exhibiting very low specific uptake were selected from a Singaporean amyloid-PET study involving 172 participants with different severities of CeVD. Two convolutional neural networks (CNN), ScaleNet and HighRes3DNet, and one conditional generative adversarial network (cGAN) were trained to map structural MR to NS PET images. NS estimates generated for all subjects using the most promising network were then subtracted from SUVr images to determine specific amyloid load only (SAβL). Associations of SAβL with various cognitive and functional test scores were then computed and compared to results using conventional SUVr. Results Multimodal ScaleNet outperformed other networks in predicting the NS content in cortical gray matter with a mean relative error below 2%. Compared to SUVr, SAβL showed increased association with cognitive and functional test scores by up to 67%. Conclusion Removing the undesirable NS uptake from the amyloid load measurement is possible using deep learning and substantially improves its accuracy. This novel analysis approach opens a new window of opportunity for improved data modeling in Alzheimer’s disease and for other neurodegenerative diseases that utilize PET imaging.

Deep learning improves utility of tau PET in the study of Alzheimer's disease

“…We additionally show that multiple generations of tau radioligands can be integrated into a single framework 7,11 . Earlier deep learning efforts with tau PET have sought to simplify 20 and augment 17 preprocessing steps; and have provided initial proof of feasibility and model interpretation 22 . The main metrics of the models compare similarly with prior deep learning classification tasks that used different sets of PET radioligands, 16,22,46 though comparisons are of course difficult given differences in available clinical outcomes.…”

“…Machine learning techniques, especially convolutional neural networks, have shown promise in a number of contexts important to neuroimaging, including with MRI 14,15 and fluorodeoxyglucose (FDG) PET. 16 Similarly, deep learning with PET radioligands has demonstrated improved research efficacy 17 and diagnostic utility. 18,19 To our knowledge, relatively few studies exist that use tau PET, [20][21][22] in part because of current limitations/challenges of applying deep learning in clinical research (especially the relatively small number of absolute subjects).…”

Artificial intelligence with deep learning in nuclear medicine and radiology

“…In this regard, DL developed for quantifying amyloid burden with increased accuracy may prove of great value. Further, as several radiotracers are available for that purpose, the approach proposed by Kang et al for translating the results obtained with [ 11 C]PIB and [ 18 F]Florbetapir into one another, appears highly attractive [203,204].…”

β-amyloid PET harmonisation across longitudinal studies: Application to AIBL, ADNI and OASIS3

“…These methods all show good correlation with standard CL or SUVR, while improving the separation between Healthy Controls (HC) and AD patients ( Pegueroles et al, 2021 ; Whittington and Gunn, 2019 ), increasing the correlation with cognitive measures ( Liu et al, 2021 ) and reducing longitudinal variability ( Bourgeat et al, 2021 ; Whittington and Gunn, 2019 ). These methods include Non-negative Matrix Factorisation (NMF) ( Bourgeat et al, 2021 ), AmyQ ( Pegueroles et al, 2021 ) and A β -index ( Leuzy et al, 2020 ) which both rely on a PCA decomposition, Amyloid Load (Amyloid IQ ) ( Whittington and Gunn, 2019 ) which uses an image-base regression, and a more recent deep-learning based method which learns to separate the specific from the non-specific binding based on A β - scans ( Liu et al, 2021 ). To our knowledge, our previous work on NMF was the only approach to explicitly enforce consistency between the decomposition of each tracer, and attempt to implicitly reduce the variability due to the use of different scanners.…”