[P06.034] The Power of Multimodal Neuroimaging Biomarkers for Clinical Trial Screening

Corey McMillan, Philadelphia, PA, Brian Avants, Philadelphia, PA, Lyle Ungar, Philadelphia, PA, John Q. Trojanowski, Philadelphia, PA, Murray Grossman, Philadelphia, PA

OBJECTIVE: To evaluate the minimum sample size (MSS) required to predict underlying pathology associated with Alzheimer's disease (AD) or frontotemporal lobar degeneration (FTLD) using MRI measures of grey matter (GM) and diffusion tensor imaging (DTI) measures of white matter (WM). BACKGROUND: As disease-modifying agents emerge for clinical trials it is critical to establish robust and powerful biomarkers for predicting AD or FTLD pathology in individual patients. The identification of statistically powerful biomarkers can achieve substantial cost-savings and increase feasibility of clinical trials involving rare disorders. Recent work suggests that both MRI and DTI can reliably predict AD or FTLD pathology, however the MMS required using these methods has not been evaluated. DESIGN/METHODS: 139 patients clinically-diagnosed with AD or FTD underwent lumbar puncture, MRI, and DTI scans. Pathology was defined using an autopsy-validated CSF surrogate of total-tau to beta-amyloid ratio: AD (N=50) and FTLD (N=89). MRI and DTI images were processed using Eigenanatomy, a statistically robust dimensionality reduction tool that identifies volumes of interest (VOIs) that account for the greatest variance in a dataset. The imaging cohort was randomly divided into train and test datasets. Within the training dataset we used linear regression and cross-validation with 1000 permutations to identify VOIs that reliably achieve accurate prediction. We then applied these VOIs to the test cohort to evaluate prediction accuracy to determine MSS estimates to achieve beta=0.8. RESULTS: We observed that multimodal neuroimaging achieved higher prediction accuracy (88%) relative to MRI (72%) and DTI (81%) alone. Power analyses suggested MTI and DTI together require the smallest MMS (N=29) to predict pathology in comparison to DTI (N=58) or MRI (N=48) alone. CONCLUSIONS: Multimodal neuroimaging biomarkers that integrate measures of WM and GM provide a statistically powerful method for predicting underlying pathology in order to screen patients for clinical trials. Supported by: The Wyncote Foundation and NIH: NS065347, AG010124, AG032953, AG017586, AG000255, NS044266, AG015116, NS053488.
Category - Aging and Dementia: Imaging

Thursday, March 21, 2013 7:30 AM

Session P06: Positron Emission Tomography and Other Imaging (7:30 AM-12:00 PM)