HEAD MOTION DEGRADES MACHINE LEARNING CLASSIFICATION OF ALZHEIMER’S DISEASE FROM POSITRON EMISSION TOMOGRAPHY - Abstract

Sep 17, 2026

Uploaded by Theodora

Brain positron emission tomography (PET) imaging is used in research and clinical practice to study, diagnose, and stage Alzheimer’s disease (AD), yet its full potential is limited by the lack of portable motion-correction solutions. This work shows that head motion during acquisition not only degrades image quality and biases tracer uptake quantification, but also affects machine-learning AD classification. A binary PET-image classifier is proposed, achieving high accuracy when motion is corrected and substantially dropping accuracy without correction, confirmed using two relevant tracers.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages5

Reading Time13 min

AccessFree to download

Abstract

Key problem: motion affects PET quality and ML classification

Proposed method: binary ML from PET images

Validation: two tracers and quantified accuracy drops

Implication: need for motion correction in PET-based ML

HEAD MOTION DEGRADES MACHINE LEARNING CLASSIFICATION OF ALZHEIMER’S DISEASE FROM POSITRON EMISSION TOMOGRAPHY Ele´onore V. Lieffrig 1, Takuya Toyonaga2, Jiazhen Zhang 1, John A. Onofrey 1 ,2 ,3 Departments of 1Biomedical Engineering, 2Radiology and Biomedical Imaging and 3 Urology, Yale University, New Haven, CT, USA arXiv :2501 .08459v1 [ ee ss .IV] 14 Jan 2025 ABSTRACT Brain positron emission tomography (PET) imaging is broadly used in research and clinical routines to study, diagnose, and stage Alzheimer’s disease (AD) . However, its full potential cannot be fully exploited yet due to the lack of portable motion correction solutions, especially in clinical settings. Head motion during data acquisition has indeed been shown to degrade image quality and induces tracer uptake quantification error. In this study, we demonstrate that it also biases machine learning-based AD classification. We start by proposing a binary classification algorithm solely based on PET images. We find that it reaches a high accuracy in classifying motion corrected images into cognitive normal or AD. We demonstrate that the classification accuracy substantially decreases when images lack motion correction, thereby limiting the algorithm’s effectiveness and biasing image interpretation. We validate these findings in cohorts of 128 11 C-UCB-J and 173 18F-FDG scans, two tracers highly relevant to the study of AD. Classification accuracies decreased by 10% and 5% on 20 18F-FDG and 20 11 C-UCB-J testing cases, respectively. Our findings underscore the critical need for efficient motion correction methods to make the most of the diagnostic capabilities of PET-based machine learning. Index Terms— Alzheimer’s disease, Brain, PET, Motion Correction, Classification, Supervised Learning, ResNet, SVM, 11 C-UCB-J, 18F-FDG 1. INTRODUCTION Alzheimer’s disease (AD) is a neurodegenerative condition characterized by progressive loss of memory and thinking abilities. AD and related dementia affect 50 million people worldwide and their prevalence is expected to reach 152 million individuals by 2050 in the absence of effective prevention or treatment [1], which makes it a major public health concern. Brain positron emission tomography (PET) imaging plays a pivotal role in AD diagnosis and staging in clinical routines. It has the potential to allow for early detection of AD, due to its ability to highlight disease features up to decades before the onset of clinical symptoms [2] . In addition, it has a crucial role in deepening our un- Axial plane 18 F-FDG 11 C-UC B-J No motion Motion No motion Motion correction cor rected correction corrected Fig. 1. (Top) Comparison of 11 C-UCB-J and 18F-FDG PETscans reconstructed with and without motion correction in patients with AD. (Bottom) Example of 11 C-UCB-J AD patient misclassified as cognitive normal in the absence of motion correction. derstanding of AD through imaging-based biomarkers and supporting the development of treatment. In recent years, machine learning (ML) based on non-deep learning, e.g. support vector machine (SVM) of region of interest (ROI) uptake [3], and deep learning [4, 5] techniques have been applied to classify cognitive normal (CN), mild cognitive impairment (MCI), and different AD stages from PET studies. These classification algorithms typically involve collection of multimodal data [6] comprising PET images (tau, amyloidor FDG-PET), neuropsychological evaluation scores (e.g. MMSE), and demographic information that can be part of the publicly-available Alzheimer’s Disease Neuroimaging Initiative (ADNI) resources. They commonly involve cumbersome image preprocessing to compute tracer standardized uptake value ratio (SUVR) for specific brain regions of interest (ROIs), which requires the co-registration of magnetic resonance imaging (MRI) with PET to segment the brain into these ROIs. Many of these methods have achieved high classification accuracy scores [6] [7] [8] . Howe

FAQs

What impact does head motion have on PET-based Alzheimer’s analysis?

Head motion degrades image quality and biases tracer uptake quantification, which can distort both diagnostic interpretation and downstream analysis.

How does the proposed method classify AD and cognitive normal using PET images?

The study proposes a binary classification algorithm based solely on PET images to separate motion-corrected images into cognitive normal or AD.

What happens to classification accuracy when motion correction is not applied?

Classification accuracy substantially decreases without motion correction, limiting algorithm effectiveness and biasing image interpretation, with reported drops across testing cases for two tracers.

Sep 17, 2026

Uploaded by Theodora