Prediction of Amyloid-β Positivity Using QSM Images Based on Bootstrap Your Own Latent

Tsubasa Kunieda, Ren Togo, Noriko Nishioka, Yukie Shimizu, Shiro Watanabe, Kenji Hirata, Keisuke Maeda, Takahiro Ogawa, Kohsuke Kudo, Miki Haseyama

Abstract: This paper presents a method to predict amyloid-$\beta$ positivity using axial slices of Quantitative Susceptibility Mapping (QSM) in Magnetic Resonance Imaging (MRI). Amyloid$-\beta$ deposition is used as a biomarker in the early diagnosis of Alzheimer’s disease by amyloid Positron Emission Tomography (PET). However, there remain concerns about low availability, radiation exposure, and cost. On the other hand, QSM-based MRI inspection does not have these disadvantages compared with PET. The proposed method provides a preliminary step before performing amyloid PET. Although our dataset used in this method is limited, we use Bootstrap Your Own Latent, which obtains highly accurate feature representations even in such cases, to achieve the effective prediction. Experimental results show that our method can predict the diagnoses of amyloid PET from QSM.