Florbetapir Image Analysis for Alzheimer's Disease Diagnosis

Abstract

Over decades Alzheimer's disease (AD) remains without decent cure, and only disease-modifying methods are available. This paper is devoted to the analysis of amyloid-PET images with florbetapir (18F-AV-45) tracer to detect the presence of AD or Mild Cognitive Impairment (MCI). The first part of the article dedicated to image processing pipeline, specifically, spacial normalisation and feature extraction. The second part is devoted to the development of the multiclass classifier with deep learning methods. In particular, deep neural network was developed to distinguish three stages: health control (HC), MCI and AD. After tuning and training a neural network, the final specificity of 78% and sensitivity of 90% has been achieved.