Semi-supervised multimodal relevance vector regression improves cognitive performance estimation from imaging and biological biomarkers

Abstract

Accurate estimation of cognitive scores for patients can help track the progress of neurological diseases. In this paper, we present a novel semi-supervised multimodal relevance vector regression (SM-RVR) method for predicting clinical scores of neurological diseases from multimodal imaging and biological biomarker, to help evaluate pathological stage and predict progression of diseases, e.g., Alzheimer's diseases (AD). Unlike most existing methods, we predict clinical scores from multimodal (imaging and biological) biomarkers, including MRI, FDG-PET, and CSF. Considering that the clinical scores of mild cognitive impairment (MCI) subjects are often less stable compared to those of AD and normal control (NC) subjects due to the heterogeneity of MCI, we use only the multimodal data of MCI subjects, but no corresponding clinical scores, to train a semi-supervised model for enhancing the estimation of clinical scores for AD and NC subjects. We also develop a new strategy for selecting the most informative MCI subjects. We evaluate the performance of our approach on 202 subjects with all three modalities of data (MRI, FDG-PET and CSF) from the Alzheimer's Disease Neuroimaging Initiative (ADNI) database. The experimental results show that our SM-RVR method achieves a root-mean-square error (RMSE) of 1.91 and a correlation coefficient (CORR) of 0.80 for estimating the MMSE scores, and also a RMSE of 4.45 and a CORR of 0.78 for estimating the ADAS-Cog scores, demonstrating very promising performances in AD studies.

Fig. 1

A template with 93 manually-labeled ROI regions used for automated labeling of MRI and PET images

Fig. 2

The flowchart of M-RVR

Fig. 3

Distributions of AD, NC and MCI subjects with CSF features. X is CSF Aβ₄₂, Y is CSF t-tau, and Z is CSF p-tau

Fig. 4

The flowchart of estimating clinical scores for MCI subjects

Fig. 5

The flowchart of M-RVR based recursive sample selection

Fig. 6

Scatter plots of actual MMSE scores vs. the estimated MMSE scores for seven different combinations of modalities. a MRI, b CSF, c PET, d MRI + CSF, e MRI + PET, f CSF + PET, and g MRI + CSF + PET

Fig. 7

Scatter plots of actual ADAS-Cog scores vs. the estimated ADAS-Cog scores for seven different combinations of modalities. a MRI, b CSF, c PET, d MRI + CSF, e MRI + PET, f CSF + PET, and g MRI + CSF + PET

Fig. 8

MMSE estimation results with respect to different combining weights of MRI, PET and CSF for SM-RVR. CSF weight denotes asC_CSF, and MRI weight denotes as C_MRI. Note that PET weight C_PET is not shown, since it can be determined as C_PET = 1 − C_CSF − C_MRI. a RMSE for MMSE Estimation, a CORR for MMSE Estimation

Fig. 9

ADAS-Cog estimation results with respect to different combining weights of MRI, PET and CSF for SM-RVR. CSF weight denotes asC_CSF, and MRI weight denotes asC_MRI. Note that PET weight C_PET is not shown, since it can be determined as C_PET = 1 − C_CSF − C_MRI. a RMSE for ADAS-Cog Estimation, b CORR for ADAS-Cog Estimation

Fig. 10

Comparison of SM-RVR with UNSM-RVR and M-RVR, with respect to the use of different number of unlabeled samples (MCI subjects). a RMSE in estimating MMSE, b CORR in estimating MMSE, c RMSE in estimating ADAS-Cog, and d CORR in estimating ADAS-Cog