This R21 resubmission application is on improving the accuracy of activation detection using functional Magnetic Resonance Imaging (fMRI). Over the past two decades this imaging modality has evolved into a noninvasive tool for understanding human cognitive and motor functions. Data collection followed by data analysis produces an activation map that highlights voxels, or volume elements, where there is brain activity in response to a stimulus or task (a paradigm). Unfortunately, the experimental data can vary greatly because of scanner variability, potential inherent unreliability of the MR signal, between-subject variability, subject motion or the several-seconds delay in the onset of the MR signal as a result of the passage of the neural stimulus through the hemodynamic lter. The result can be vast differences in activation maps from one scanning session to the next, even when the same subject is administered the same paradigm. There has been much recent work to assess reliability of activation maps in multiple settings. Many have incorporated results on multiple hypothesis tests in a somewhat post hoc manner to improve the reliability and consistency in activation detection. To account for the fact that activated voxels tend to occur in clusters, a common approach incorporates the Ising model, from statistical physics, where each voxel is either activated or not, but with some dependence on the states of its neighbors. Almost no methods take advantage of the well-known belief that only 2-3% of the voxels are truly active in a typical fMRI experiment, and no method has yet incorporated both this expectation on the proportion of activated voxels and the spatial context. Requiring exactly 2-3% activated voxels in the activation maps is not an accurate representation of our prior knowledge that 2-3% of voxels are activated on average and would increase the chance of missing pathologies and hence mis-diagnosing anomalies in a clinical setting. This proposal explores new approaches to improving activation detection by constraining the parameters of the Ising model so the a priori expected proportion of truly active voxels is restricted to the desired range.
The specific aims proposed are: 1) to investigate approaches to specify the expected proportion of activated voxels in the Ising model to be the a priori value and 2) to develop a computationally practical approach to estimate the model parameters and produce activation maps in the context of the complexities introduced in 1). Our proposal will allow inclusion of researcher uncertainty about the constraint and anatomic information in the spatial context. Each e ort is specifically motivated and will contribute, if successful, to the development of reliably consistent within-subject fMRI activation maps and also to identify anomalies in activation across subjects. A range of data from realistic computer simulations and archived human data on motor task experiments and working memory experiments in traumatic brain injury (TBI) patients and normal subjects will be used to explore, develop and re ne the suggested approaches. Open-source software, along with detailed tutorials on best practices and pitfalls, will also be developed and made available in order to facilitate early adoption by practitioners in fMRI. 1
This project aims to develop statistical and computational methodology for improved detection of activation in functional Magnetic Resonance Imaging (fMRI) studies. If successful, the proposed activity will make it possible to obtain more reliable activation maps, and thus, provide investigators with the ability to identify pathologies and anomalous activation patterns. In the long term, the proposed activity has the potential for wider adoption of fMRI as a tool for research and also for diagnostic purposes. 1
| Adrian, Daniel W; Maitra, Ranjan; Rowe, Daniel B (2018) COMPLEX-VALUED TIME SERIES MODELING FOR IMPROVED ACTIVATION DETECTION IN FMRI STUDIES. Ann Appl Stat 12:1451-1478 |
| Peterson, Anna D; Ghosh, Arka P; Maitra, Ranjan (2018) Merging K-means with hierarchical clustering for identifying general-shaped groups. Stat (Int Stat Inst) 7: |
| Adrian, Daniel W; Maitra, Ranjan; Rowe, Daniel B (2013) Ricean over Gaussian modelling in magnitude fMRI Analysis-Added Complexity with Negligible Practical Benefits. Stat 2:303-316 |
