Deep learning for FDG-PET classification in patients with Alzheimer’s disease, dementia with Lewy bodies and their mixed pathology: a solution for diagnostic heterogeneity

TL;DR

This study uses deep learning on FDG-PET scans to improve diagnosis of Alzheimer’s, dementia with Lewy bodies, and mixed cases, which are often hard to distinguish.

Contribution

A novel deep learning model is proposed to classify AD, DLB, mixed pathology, and healthy controls using FDG-PET data.

Findings

The model achieved an AUROC of 0.90 for DLB classification.

FDG-PET was shown to be a useful biomarker for differentiating AD, DLB, mixed, and healthy groups.

The model outperforms existing methods in classifying these complex neurological conditions.

Abstract

Mixed pathology of Alzheimer’s disease (AD) and dementia with Lewy bodies (DLB) are frequently observed in patients with cognitive impairment, and complicate clinical diagnosis. We aimed to develop a classification model using 18F-fluorodeoxyglucose (FDG) positron emission tomography (PET) to improve diagnostic accuracy for these challenging cases.

We analyzed FDG-PET images from 277 participants who were categorized into AD, DLB, mixed disease, and healthy control (HC) groups. Deep learning-based classification models were trained on seven binary classification tasks and one multiclass classification task and subsequently integrated into an ensemble model to predict AD, DLB, mixed disease or HC groups.

The model achieved an AUROC of 0.73 (95% CI, 0.69–0.78) for AD, 0.90 (95% CI, 0.89–0.91) for DLB, 0.71 (95% CI, 0.66–0.75) for Mixed, and 0.87 (95% CI, 0.84–0.89) for HC.

The model represents the state-of-the-art in automatic FDG-PET-based classification of AD, DLB, Mixed, and HC. This study highlights the utility of FDG-PET as a biomarker for differentiating AD, DLB, Mixed, and HC groups, resolving diagnostic challenges caused by overlapping clinical features.

Figures