MRI-free amyloid PET quantification using a deep learning model and white matter reference
Seung-Ho Shin1, Yongcheol Mo2, Hee Sung Hwang3
1Department of Biomedical Informatics, Hallym University Chuncheon Sacred Heart Hospital, Chuncheon-si, Republic of Korea.
Journal of Alzheimer'S Disease : JAD
|July 23, 2026
Summary
A new deep learning model accurately segments brain tissues from amyloid PET scans, enabling MRI-free Alzheimer's disease diagnosis. This method simplifies workflows while maintaining diagnostic accuracy for Alzheimer's disease (AD).
Area of Science:
- Neuroimaging
- Artificial Intelligence in Medicine
- Alzheimer's Disease Diagnostics
Background:
- Accurate Standardized Uptake Value Ratio (SUVR) quantification in amyloid Positron Emission Tomography (PET) is crucial for Alzheimer's disease (AD) diagnosis.
- Traditional methods require Magnetic Resonance Imaging (MRI) for gray and white matter segmentation due to subtle uptake differences.
- This reliance on MRI complicates the clinical workflow for AD diagnosis.
Purpose of the Study:
- To develop and validate a 3D deep learning model for segmenting gray and white matter directly from PET images.
- To enable MRI-free SUVR quantification for improved Alzheimer's disease diagnosis.
- To assess the diagnostic performance of the MRI-free approach.
Main Methods:
- A retrospective study included 373 participants (training/test sets) and external validation with 625 PET/CT scans (Alzheimer's Disease Neuroimaging Initiative).
- A 3D deep learning model was trained to segment gray matter (GM) and white matter (WM) from amyloid PET images.
- Model performance was evaluated using Dice coefficients and Intersection over Union (IoU); SUVRs were compared to MRI-based references using Spearman correlation and ROC analysis.
Main Results:
- The deep learning model achieved high segmentation accuracy: Dice coefficients of 0.785 (GM) and 0.838 (WM) internally, and 0.743 (GM) and 0.803 (WM) externally.
- PET/CT-derived SUVRs strongly correlated with MRI references (Spearman's ρ ≥ 0.98, p < 0.001).
- SUVRs derived from the model predicted amyloid status (AUC 0.86 for SUVRGM, 0.85 for SUVRGM/WM) and cognitive impairment (AUC 0.78).
Conclusions:
- Deep learning-based segmentation of amyloid PET images enables accurate, MRI-free quantification of GM and WM SUVRs.
- This simplified approach maintains diagnostic performance comparable to traditional MRI-based methods for AD diagnosis.
- The model offers a streamlined workflow for clinical application in Alzheimer's disease assessment.
