5️⃣ Big thanks to all co-authors! www.medrxiv.org/content/10.6...
3️⃣ Synthetic scans showed high potential in clinically relevant tasks such as predicting progression to dementia in cognitively unimpaired individuals.
4️⃣ Want to try it out yourself? The trained model and project code is openly available on GitHub for research use (License: CC BY-NC)! 🐙
This study was led by María Rivera Sánchez. A huge thanks to all co-authors: @rikossenkoppele.bsky.social, and those not on bsky.
📢 New preprint from BioFINDER, led by
@karlssonlinda1.bsky.social!
We trained a deep learning model to generate synthetic tau-PET 3D scans from more accessible data: structural MRI, plasma p-tau217, and age. medrxiv.org/content/10.6...
2️⃣ Synthetic scans showed high agreement with true scans regarding both tau load and distribution. Performance was highest when using only MRI, age and plasma p-tau217 as input, but also strong with only MRI and age.
1️⃣ We assembled data from 13 Alzheimer's disease cohorts covering a broad spectrum of clinical diagnoses (~5000 individuals) and systematically optimized a 3D U-Net model, evaluated on held-out test data.
New paper in Neurology! We investigated whether future AD dementia can be predicted in individuals with subjective cognitive decline. We found that plasma p-tau217, cognitive testing, and APOE ε4 status accurately identify those at highest risk. neurology.org/doi/pdf/10.1...
