Multi-modal multi-task learning for joint prediction of clinical scores in Alzheimer's disease

Authors:
Daoqiang Zhang;Dinggang Shen
Affiliations:
Dept. of Radiology and BRIC, University of North Carolina at Chapel Hill, NC and Dept. of Computer Science and Engineering, Nanjing University of Aeronautics and Astronautics, Nanjing, China;Dept. of Radiology and BRIC, University of North Carolina at Chapel Hill, NC
Venue:
MBIA'11 Proceedings of the First international conference on Multimodal brain image analysis
Year:
2011

Citing 5
Cited 0

Heterogeneous data fusion for alzheimer's disease study

Proceedings of the 14th ACM SIGKDD international conference on Knowledge discovery and data mining
MKL for Robust Multi-modality AD Classification

MICCAI '09 Proceedings of the 12th International Conference on Medical Image Computing and Computer-Assisted Intervention: Part II
Multi-task feature learning via efficient l2, 1-norm minimization

UAI '09 Proceedings of the Twenty-Fifth Conference on Uncertainty in Artificial Intelligence
Joint estimation of multiple clinical variables of neurological diseases from imaging patterns

ISBI'10 Proceedings of the 2010 IEEE international conference on Biomedical imaging: from nano to Macro
Predicting clinical variable from MRI features: application to MMSE in MCI

MICCAI'05 Proceedings of the 8th international conference on Medical Image Computing and Computer-Assisted Intervention - Volume Part I

Quantified Score

Hi-index	0.00

Visualization

Abstract

One recent interest in computer-aided diagnosis of neurological diseases is to predict the clinical scores from brain images. Most existing methods usually estimate multiple clinical variables separately, without considering the useful correlation information among them. On the other hand, nearly all methods use only one modality of data (mostly structural MRI) for regression, and thus ignore the complementary information among different modalities. To address these issues, in this paper, we present a general methodology, namely Multi-Modal Multi-Task (M3T) learning, to jointly predict multiple variables from multi-modal data. Our method contains three major subsequent steps: (1) a multi-task feature selection which selects the common subset of relevant features for the related multiple clinical variables from each modality; (2) a kernel-based multimodal data fusion which fuses the above-selected features from all modalities; (3) a support vector regression which predicts multiple clinical variables based on the previously learnt mixed kernel. Experimental results on ADNI dataset with both imaging modalities (MRI and PET) and biological modality (CSF) validate the efficacy of the proposed M3T learning method.