DeepAD: A deep learning application for predicting amyloid standardized uptake value ratio through PET for...

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Biomarker Discovery

Background:

• Amyloid deposition is a key biomarker for Alzheimer's disease diagnosis.
• ¹⁸F-florbetapir PET scans are used to quantify cortical amyloid.
• Current methods for amyloid quantification are labor-intensive and require specialized resources, limiting accessibility.

Purpose of the Study:

• To develop and validate a deep learning model for efficient and accurate amyloid deposition quantification.
• To assess the performance of convolutional neural networks (CNNs) and Gradient Boosting Decision Trees (GBDTs) for predicting amyloid SUVR.
• To create an accessible tool for Alzheimer's diagnosis support.

Main Methods:

• Utilized data from the Alzheimer's Disease Neuroimaging Initiative (ADNI) including PET imaging, clinical, and genetic data from 2,980 patients.
• Evaluated various CNN architectures (ResNet, EfficientNet, RegNet) and combined the best performing model with GBDT for regression.
• Optimized model configurations for predicting standardized uptake value ratio (SUVR) of amyloid.

Main Results:

• The RegNet X064 architecture combined with a grid search-tuned GBDT achieved the lowest loss.
• The model reached a Mean Absolute Error (MAE) of 0.0441, corresponding to 96.4% accuracy on a 596-patient test set.
• The deep learning approach demonstrated higher consistency and lower margins of error compared to human readers.

Conclusions:

• The developed deep learning model offers a more consistent, accessible, and faster method for amyloid quantification.
• The web application DeepAD makes this diagnostic tool widely accessible, particularly for resource-limited settings.
• This approach shows potential for broader applications in medical imaging analysis and Alzheimer's disease research.