MRI Region of Interest Definitions
Autism & Fragile X Longitudinal Studies of Brain Development
UNC NeuroImage Analysis Lab
The UNC pediatric longitudinal MRI studies of brain development in Autism and Fragile X utilize measurement protocols that are either adapted from existing protocols in use by other research groups, or developed by our own group. As participants in the UNC NeuroImage Analysis Lab, we have many collaborative efforts with other research teams in developing protocols. Protocols developed by our own group are guided by methods described in the literature, available methods for similar measurements, classic brain atlases, and expert pediatric neuroradiologists. When adapting or developing protocols, our group is confronted with the challenges of pediatric brain images. Protocols from research groups studying adults often require at least some adaptation to accommodate pediatric images; the level of adaptation required depends upon the complexity of the brain region of interest and the timeline of brain development.
Each Region of Interest measurement protocol indicates major boundaries of the region, specific tracing guidelines, tool used for defining the region, approximate time required for completion of the protocol, and established rater reliability.
Processing Tools
Several tools are employed in the measurement of MRI data in the UNC NeuroImage Analysis Lab, ranging from fully automated (e.g., EMS for tissue segmentation) to semi-automated (e.g., MOJO for hippocampal segmentation) to fully manual tools (e.g., ITK-SNAP for amygdala segmentation).
A core tool for user-supervised segmentation is ITK-SNAP, which allows manual tracing in all three planes with 2-D overlay and 3D visualization, as well as semi-automated 3D snake segmentation via 3D geodesic level sets. ITK-SNAP (InsightSNAP) and other tools are freely available through the NeuroLib Software web site.
Reliability Procedures
The UNC NeuroImage Analysis Lab takes great care in developing valid, reliable, and reproducible volumetric measurement protocols for various brain regions. Our standard procedure for establishing reliability is to have at least two raters perform the measurement protocol on a set of 15 images, which includes 5 images repeated three times in random order. Intra-rater reliability of greater than 0.90 is required for a rater to be deemed reliable. Requirements for inter-rater reliability vary based on difficulty and size of structure, but are typically greater than 0.90 for large structures or when two raters will be sharing the workload, or greater than 0.80 for smaller structures, or when only a single rater will be tracing the full data set. For studies of pediatric subjects, we always establish rater reliability on pediatric brain images from own dataset.