The study used longitudinal visit data from multiple modalities (MRI, cognition, and clinical data) to predict the risk of progression from mild cognitive impairment (MCI) to Alzheimer's disease (AD) at each timepoint in the visit trajectory.
The key highlights and insights are:
The proposed hierarchical multi-task framework first learned to predict a set of neuropsychological composite cognitive function scores as auxiliary tasks. The weighted combination of these forecasted auxiliary task scores was then used to predict the risk of progression from MCI to AD.
The hierarchical structure allowed better optimization and knowledge transfer between the main and auxiliary tasks, leading to improved performance compared to traditional multi-task and single-task baselines.
The relevance weights assigned to the forecasted composite scores provided explanations about the potential cognitive factors contributing to the risk of progression at each timepoint. This allowed the model to inform clinicians 6 months in advance about the specific cognitive declines that may lead to future progression.
Ablation studies demonstrated that imaging and cognition data were the most discriminative features for predicting progression, while clinical data alone had lower predictive performance.
The proposed framework is flexible and can be generalized to other clinical applications beyond Alzheimer's disease by incorporating additional modalities and selecting appropriate auxiliary tasks.
To Another Language
from source content
arxiv.org
Дополнительные вопросы