Predicting brain activity using a Bayesian spatial model

Abstract

Increasing the clinical applicability of functional neuroimaging technology is an emerging objective, e.g. for diagnostic and treatment purposes. We propose a novel Bayesian spatial hierarchical framework for predicting follow-up neural activity based on an individual's baseline functional neuroimaging data. Our approach attempts to overcome some shortcomings of the modeling methods used in other neuroimaging settings, by borrowing strength from the spatial correlations present in the data. Our proposed methodology is applicable to data from various imaging modalities including functional magnetic resonance imaging and positron emission tomography, and we provide an illustration here using positron emission tomography data from a study of Alzheimer's disease to predict disease progression.

Keywords: Alzheimer's disease; Bayesian spatial modeling; neuroimaging; prediction.

Figure 1

Individualized predicted and observed month 6 follow-up regional glucose uptake measurements for 4 AD patients from the test data set. Axial slice 40 is shown in radiological view. There is a satisfactory agreement between the observed and predicted post-treatment regional glucose uptake.

Figure 2

Individualized predicted and observed month 6 follow-up regional glucose uptake measurements for 4 HC subjects from the test data set; axial slice 40 is shown in radiological view. There is a satisfactory agreement between the observed and predicted post-treatment regional glucose uptake.

Figure 3

The images depict the square root of the prediction mean square error, divided by the average observed brain activity (stPMSE, (2)) at each voxel for prediction of the follow-up activity for 33 test subject in the AD group. Axial slices 35, 40 and 45 are shown (in radiological view) of the stPMSE map based on (a) our proposed model, (b) BHM proposed in [1], (c) BSMac proposed in [4], and (c) GLM. Average errors: 0.083 for the BSPM (a), 0.080 for the BHM (b) 0.104 for the BSMac (c), and 0.156 for the GLM (d).

Figure 4

Region functional connectivity for AD (top) and HC (bottom) subjects. The regions that have (posterior median) correlations exceeding 0.75 are shown. The connecting lines have different thicknesses, corresponding to the strength of the inter-regional correlations.