A Modular Framework to Predict Alzheimer’s Disease Progression Using Conditional Generative Adversarial Networks

摘要

Alzheimer’s disease (AD) is a chronic neurodegenerative disease that worsens over time. The number of AD cases is growing, around 3 million new US cases each year. Although state-of-the-art research shows promise, predicting the disease’s rate of progression for a case by case basis remains a challenging problem. Current methods of predicting the progression of AD can delay treatment and lead to misdiagnosis. We propose a novel approach to simulate the rate of progression of AD and the atrophy of the brain over time. We seek to achieve this by generating synthetic magnetic resonance (MR) images via a series of Conditional Deep Convolutional Generative Adversarial Neural Networks (CDCGANs) and then analyze them by computing the fractal dimensionality of the cortical brain ribbons. This paper shows the feasibility of this proposal by cascading CDCGANs that simulate different stages of AD. It is possible to extend by a tandem of CDCGANs that would simulate the different stages of the disease. MR images used here are from ADNI(Alzheimer’s Disease Neuroimaging Initiative). The atrophy is measure using fractal dimension (box-counting method)of the cortical ribbon(CR). A decreasing fractal dimension is a confirmation that the disease progress over time.