Outlier Detection for Diffusion Tensor Imaging by Testing for ADC Consistency

¹Johns Hopkins University, Baltimore, MD, USA; ²Kennedy Krieger Institute, Baltimore, MD, USA

Diffusion tensor imaging (DTI) is an important tool to study brain white matter anatomy and its abnormalities. However, DTI-derived variables are affected by various sources of signal uncertainty. In this work, a new method to automatically detect and remove corrupted diffusion-weighted images due to subject motion is proposed. This approach iteratively identifies potential outliers by evaluating error-maps created from the apparent diffusion constant (ADC) maps derived from the original diffusion-weighted images and the estimated tensor matrix. Error clustering and gradient-neighboring analyses were used as criteria for outlier judgment.