[Paper Review] MRI-based Multi-task Decoupling Learning for Alzheimer's Disease Detection and MMSE Score Prediction: A Multi-site Validation

Xu Tian, Jin Liu+4 more|arXiv (Cornell University)|Apr 2, 2022

Brain Tumor Detection and Classification4 citations

TL;DR

This paper proposes a MRI-based multi-task decoupled learning framework for Alzheimer's disease detection and MMSE score prediction, leveraging shared feature representations through three multi-task interaction layers with feature decoupling and consistency loss to enhance generalization. The method outperforms single-task and existing state-of-the-art approaches on multi-site MRI datasets.

ABSTRACT

Accurately detecting Alzheimer's disease (AD) and predicting mini-mental state examination (MMSE) score are important tasks in elderly health by magnetic resonance imaging (MRI). Most of the previous methods on these two tasks are based on single-task learning and rarely consider the correlation between them. Since the MMSE score, which is an important basis for AD diagnosis, can also reflect the progress of cognitive impairment, some studies have begun to apply multi-task learning methods to these two tasks. However, how to exploit feature correlation remains a challenging problem for these methods. To comprehensively address this challenge, we propose a MRI-based multi-task decoupled learning method for AD detection and MMSE score prediction. First, a multi-task learning network is proposed to implement AD detection and MMSE score prediction, which exploits feature correlation by adding three multi-task interaction layers between the backbones of the two tasks. Each multi-task interaction layer contains two feature decoupling modules and one feature interaction module. Furthermore, to enhance the generalization between tasks of the features selected by the feature decoupling module, we propose the feature consistency loss constrained feature decoupling module. Finally, in order to exploit the specific distribution information of MMSE score in different groups, a distribution loss is proposed to further enhance the model performance. We evaluate our proposed method on multi-site datasets. Experimental results show that our proposed multi-task decoupled representation learning method achieves good performance, outperforming single-task learning and other existing state-of-the-art methods.

Motivation & Objective

• To address the limited exploitation of correlation between Alzheimer's disease detection and MMSE score prediction in existing MRI-based methods.
• To improve model generalization by explicitly decoupling shared and task-specific features using a consistency loss.
• To enhance prediction performance by modeling the distribution of MMSE scores across different cognitive impairment groups.
• To validate the method on multi-site MRI datasets to ensure robustness and clinical applicability.

Proposed method

• A multi-task learning network is designed with shared backbones for AD detection and MMSE prediction, connected by three multi-task interaction layers.
• Each multi-task interaction layer includes two feature decoupling modules and one feature interaction module to model cross-task dependencies.
• A feature consistency loss is introduced to regularize the decoupled features, improving generalization across tasks.
• A distribution loss is proposed to capture group-specific MMSE score distributions, enhancing regression accuracy.
• The model is trained end-to-end on multi-site MRI datasets to ensure robustness to domain shift.

Experimental results

Research questions

• RQ1 How can shared and task-specific features be effectively decoupled to improve performance in joint AD detection and MMSE prediction?
• RQ2 To what extent does enforcing feature consistency across tasks improve model generalization in multi-site MRI settings?
• RQ3 Can modeling the distribution of MMSE scores across cognitive groups further enhance prediction performance?
• RQ4 How does the proposed method compare to single-task learning and existing state-of-the-art multi-task approaches on multi-site data?

Key findings

• The proposed multi-task decoupled learning method achieves superior performance compared to single-task learning on both AD detection and MMSE prediction tasks.
• The inclusion of feature consistency loss leads to improved generalization of shared representations across tasks.
• The distribution loss significantly enhances MMSE score prediction accuracy by capturing group-wise variations in cognitive decline.
• The model demonstrates strong robustness and generalization across multiple independent MRI datasets from different sites.