30/11/2004


___________________________________________
Overview

-Code works properly and produces many results in parallel.
-Some descriptions, results and conclusions are located at:

    o http://www.schestowitz.com/Research/Explanatory_Notes/2004/NRR_VXL/nrr_vxl_20_nov_2004.pdf
    o http://www.schestowitz.com/Research/Explanatory_Notes/2004/NRR_VXL/nrr_vxl_24_nov_2004.pdf
    o http://www.schestowitz.com/Research/Explanatory_Notes/2004/NRR_VXL/nrr_vxl_25_nov_2004.pdf
    o http://www.schestowitz.com/Research/Explanatory_Notes/2004/NRR_VXL/nrr_vxl_26_nov_2004.pdf

-Below is a summary.


___________________________________________
Some Experiments (partial list)

-(Visual) Evaluation of the performance of 3-D pair-wise registration.
-Checking different weighting for shape versus intensity.
-Trying and comparing a large variety of objective functions.
-Checking the effect of variation in number of knot-points.
-Checking the effect of different multi-resolution approaches.
-Looking at group-wise and MDL approaches.


___________________________________________
Some Results and Conclusions

-Reasonably good results can be obtained after less than half an hour per pair.
-Change in the shape-intensity weighting affects results considerably.
-Advantageous to use intensity component, less of the shape component.
-Group-wise registration appears to give better results than just pair-wise registration. It is not possible to quantify quality yet.
-Increase in number of knot-points improves registration, but not when exceeding a certain number, e.g. when a small volume is occupied by very dense knot-points.
-Increase in resolution does not improve the results, only slows down the process.


___________________________________________
Pending Experiments

-Trying large group-wise optimisations in 3-D.
-Trying algorithms recently proposed.
-Looking at even larger registration tasks, e.g. more iterations, more knot-points.
-Involvement of models and analysis of discrepancies (current work).


___________________________________________
Possible Experiments

-Finding ways to evaluate results.
-Expanding the set of 8 brain volumes, e.g. by warping and perturbing existing ones.


___________________________________________
Misc.

-MATLAB registration code was set aside.
-Need for more filestore space for more 3-D brain data.