Deep learning-based classification improves clinical interpretation of [18F]florzolotau PET

Ziel/Aim The latest developments in tau PET have made in vivo visualization of tau distribution possible. However, clinical interpretation of tau PET remains challenging due to few visual assessment guidelines and inter-reader variability. The selection of the reference region, as well as inconsistent cut-off criteria will affect the final interpretation and may underestimate actual pathological changes. Deep learning based-medical image classification has shown high potential in neuroimages during the past years. Given the difference among tau ligands and the lack of exploration of Progressive supranuclear palsy (PSP), we trained a Convolutional Neural Network (CNN) on [18F]florzolotau SUV PET images to classify AD, PSP, and controls.