Hi Chao
With respect to your second item. Your command line looks perfectly okay, the issue is with your manual segmentations.

The error in the shape is due to an issue when mapping the surface to the sphere, usually happens when parts of the surface are connected only via single edge of a voxel (diagonal connectedness) rather than a simple 6-connectedness of the segmentation. There are basically 3 ways to try to work on this issue:
a) edit your post-processed (pp) label/segmentations and make sure there are no holes, handles and diagonal connections. This can be quite time consuming
b) run the SegPostProcessCLP command manually a second time on the already post-processed data. This basically does a second iteration of the geodesic level-set smoothing which often solves this problem. Simply override the current pp-label file and remove the spharm surfaces and coefficients and then rerun ShapeAnalysisModule
c) enable Gaussian smoothing. This is a more aggressive smoothing that does not smooth along the surface normal the way the standard geodesic smoothing does, but it usually solve these issues too. The default variance of the filter is though too large, it will smooth your hippocampi far too much. I would suggest a variance value of 0.5-1.5mm isotropically. So, In the "advanced post processed segmentation" section in Slicer, enable Gaussian Filtering and set the variance to 0.75mm for x, y and z. Now if you do this, then you should really recompute everything (choose a new output directory) as one should not mix cases with geodesic and Gaussian smoothing (this would likely incur a bias into your statistical analysis). Btw, if 1mm is not enough, increase its size stepwise until things work.

Martin