Discriminating coupling between structural connectivity and functional connectivity in the brain networks of juvenile myoclonic epilepsy

摘要

Regional patterns of amyloid-beta accumulation in Alzheimer’s disease: Comparison between autoencoder and the non-negative matrix factorization (NMF) The deposition of amyloid beta protein is crucially involved in Alzheimer’s disease, which may damage neuronal cells. The most common method for distinguishing amyloid positive patients from the others is the single threshold model which considers a subject as the amyloid positive when the amyloid deposition over the whole brain exceeds a certain threshold. However, it ignores the regional deposition patterns that may play an important role in the progression of AD. Thus, in this study we investigate the regional amyloid deposition patterns using two different methods: auto-encoder (AE) with non-negative weight constraints, and non-negative matrix factorization (NMF). Both methods extract local basis patterns from data of multiple subjects, and represent the deposition pattern of the whole brain by combination of the basis patterns. While NMF is a mathematical algorithm, AE is a machine learning technique. We recruited 40 amyloid positive and 47 amyloid negative subjects from Korea University Guro Hospital. The 87 subjects had 18F-Florbetaben PET scans and were grouped through the whole-brain visual assessment by profes- sional neurologists. We computed SUVR (standardized uptake value ratios) of 68 cerebral cortical regions (Desikan-Killiany atlas) with the reference of the cerebellum after partial-volume effect correction through SPM. We then used them as inputs of AE and NMF. Both methods extracted the regional amyloid deposition patterns. However, a machine learning technique out-performed; AE captured more localized patterns than those of NMF. AE also led lower reconstruction error than NMF did, meaning better performance of AE. We also explored patterns that significantly contribute to distinguish amyloid positive patients from the others, and investigated their clinical correlation with cognitive scores including the mini-mental state examination (MMSE). This showed that the regional patterns of amyloid deposition can be used as a biomarker of the Alzheimer’s disease, and can be utilized as the hallmarks of its subtyping.