Improving the Detection of Mild Cognitive Impairment with FlowGAN: a Framework for ASL to FDG-PET Image Synthesis
Year: 2025
Type: preprint
Abstract: Background and Significance 18 F-Fluorodeoxyglucose Positron Emission Tomography (FDG-PET) is commonly used to measure regional metabolism for diagnosis and monitoring of Alzheimer’s Disease (AD). However, FDG-PET is expensive and not widely available. Regional cerebral blood flow (CBF) is coupled to regional glucose metabolism and can be imaged noninvasively using Arterial Spin Labeling (ASL) perfusion MRI. Previously, we developed FlowGAN to synthesize FDG-PET images from ASL CBF and T1w MRI for lateralization of temporal lobe epilepsy (TLE). This study aims to extend FlowGAN to AD by examining the diagnostic potential of FlowGAN PET in mild cognitive impairment (MCI). Methods We included 47 MCI and 31 cognitively unimpaired (CU) individuals. FlowGAN, a generative adversarial network (GAN) for translating T1w MRI and ASL CBF maps into FDG-PET-like images, was trained using a 12-fold cross-validation scheme. We then evaluated the synthetic PET volumes by comparing them to true PET images both in appearance and for their classification performance in distinguishing MCI from CU via a random forest (RF) model, within regions of interest (ROIs). Results Synthetic FlowGAN PET volumes showed significant structural similarity to true PET volumes (SSIM = 0.958). Moreover, the performance of the best RF models for classification of MCI versus CU were comparable between the original PET and FlowGAN PET, both when considering all ROIs (PET AUC = 0.87, 95% CI: [0.79, 0.95]; FlowGAN AUC = 0.86, 95% CI: [0.77, 0.94]) and only a subset (PET AUC = 0.84, 95% CI: [0.77, 0.94]; FlowGAN AUC = 0.83, 95% CI: [0.74, 0.92]). Conclusions FlowGAN PET performs comparably to true PET in both appearance and classification. Since ASL can be acquired as part of a routine multimodal MRI protocol that is typically performed in patients with cognitive complaints, these findings may lead to improved access to diagnosis and treatment for AD.
Source: medRxiv
Authors: Chetan Vadali, Alfredo Lucas, T. Campbell Arnold, Sudipto Dolui, Sandhitsu Das, David A. Wolk, Kathryn A. Davis, Joel M. Stein, John A. Detre
Institution: University of Pennsylvania
Language: English
Subfield: Psychiatry and Mental health
Field: Medicine
Domain: Health Sciences
Open Access status: green
OpenAlex ID: W4412168639