Faculty Bibliography

We studied the extent of the residual anatomical variability (RAV) in groups of subjects after standard spatial normalization. The normalization performance of three Magnetic Resonance (MR) data analysis tools, Freesurfer[1], Statistical Parametric Mapping (SPM)[2], and Large Deformation Diffeomorphic Metric Mapping (LDDMM)[3] was examined when four regions of interest (ROIs) in basal ganglia in each hemisphere were considered. The data set consisted of 27 T1-weighted brain scans, from 14 Parkinson's Disease (PD) subjects and 13 age-matched control subjects. As expected, smaller ROIs had reduced Dice similarity coefficients (DSCs) when computed over all subject pairs with different group sizes and registration methods. The LDDMM method had the lowest RAV of the three methods, but was the most computationally intensive. This result has major implications for group fMRI studies that utilize spatial normalization as a standard pre-processing method, and supports the use of fMRI ROI analysis methods that compute significance in each subject’s native space, especially when basal ganglia structures are involved. Major factors that affect RAV were discussed.

To study changes in neuron size, number and distribution over a wide range of animal sizes, it is necessary for us to identify the axons and myelin of each neuron in electron microscope images of nerve cross-sections. Current methods commonly in use involve manually labeling each axon, which is extremely time-consuming as a single nerve contains thousands of axons. In order to make this process more efficient, we developed a computer-assisted neuron segmentation and analysis method. First we acquired a set of sub-images with identical size and resolution using a scanning electron microscope. We then developed an algorithm which used cross-correlation to stitch the sub-images into large images containing whole neuron clusters for segmentation. We developed a second algorithm to pre-process the stitched image, then segment and individually label axons using combined morphological operations. The myelin of each neuron was also segmented using a region growing algorithm with the geometric centers of axons as seeds. The final output of our algorithm is a histogram of axon and myelin sizes. We used this method to analyze nerves from different animal species including elephant, rat and shrew [1]. The typical processing time for a 4~6 million-pixel image on a PC (1.66GHz Pentium M Processor, 1G RAM) was approximately 5 minutes. The mislabel rate (percentage of false detections plus failed detections) is currently under 10% and improving. The method was proven to be well-suited for studying the effect of animal size on axon size and number

We apply a recently developed automated brain segmentation method, FS+LDDMM, to brain MRI scans from Parkinson's disease (PD) subjects, and normal age-matched controls and compare the results to manual segmentation done by trained neuroscientists. The data set consisted of 14 PD subjects and 12 age-matched control subjects without neurologic disease and comparison was done on six subcortical brain structures (left and right caudate, putamen and thalamus). Comparison between automatic and manual segmentation was based on Dice similarity coefficient (overlap percentage), L1 error, symmetrized Hausdorff distance and symmetrized mean surface distance. Results suggest that FS+LDDMM is well-suited for subcortical structure segmentation and further shape analysis in Parkinson's Disease. The asymmetry of the Dice similarity coefficient over shape change is also discussed based on the observation and measurement of FS+LDDMM segmentation results