Fitting 3-d Imaging Data Using Orthogonal Distance Regression

Orthogonal Distance Regression methods provide meaningful least squares estimates for distance calibration problems such as are required for biomedical image processing. A major characteristic of these problems is the prevalence of non-uniformly distributed noise in the data to be fitted. A NIST developed software package, ODRPACK, specifically designed for these circumstances, has been successfully employed to identify parameters and characteristics of such biotechnology models.

This talk will discuss how Image Guided Technologies (IGT) in Boulder, CO, used ODRPACK to locate significant calibration discrepancies within their coordinate measuring machine. This instrument is employed in the design and manufacture of 3D optical localizers that enable surgeons to track the location of a probe inserted into a patient's skull. The development of novel optical property models was efficiently facilitated by application of ODRPACK, and accuracy testing and certification of new IGT systems are presently performed using procedures based on ODRPACK. Some specific features of ODRPACK that make it especially relevant for high precision measurements and uniquely well-suited for modeling 3D image data will be described.