Abstract
Longitudinal analysis of a disease is an important issue to understand its progression and design prognosis and early diagnostic tools. From the longitudinal images where data is collected from multiple time points, both the spatial structural information and the longitudinal variations are captured. The temporal dynamics are more informative than static observations of the symptoms, particularly for neurodegenerative diseases such as Alzheimer's disease, whose progression spans over the years with early subtle changes. In this paper, we propose a new generative framework to predict the lesion progression over time. Our method first encodes images into the structural and longitudinal state vectors, where interpolation or extrapolation of feature vectors in the time axis can be performed for the manipulation of these feature vectors. These processed feature vectors can be decoded into image space to predict the image at the time point which we are interested in. During the training, we force the model to encode longitudinal changes into longitudinal state features and capture the structural information in a separate vector. Moreover, we introduce a personalized memory for the online update scheme, which adapts the model to the target subject, which helps the model preserve fine details of brain image structures in each subject. Experimental results on the public longitudinal brain magnetic resonance imaging dataset show the effectiveness of the proposed method.
Original language | English |
---|---|
Pages (from-to) | 143212-143221 |
Number of pages | 10 |
Journal | IEEE Access |
Volume | 9 |
DOIs | |
State | Published - 2021 |
Keywords
- Brain MR images
- deep learning
- generative model
- longitudinal analysis
- memory network
- personalized prediction