This thesis studies various aspects of the accuracy issue in image-guided neurosurgery (). Factors such as registration, MR geometrical distortions and accuracy of digitizing device, are theoretically described in a general model of .
The means of registering the patient to its image data is then studied, starting with the definition of registration error. Computer simulations of registration by homologous point matching are described along with a clinical study comparing homologous point matching and surface matching registration methods.
A 3-D MR geometrical distortion experiment performed on a stereotactic frame is presented. These measurements demonstrate the discrepancy that can be observed in the geometry of the frame when imaged with differing read-out gradient directions, and quantitatively evaluate the geometrical distortion associated with the image of the frame of known geometry.
Since geometrical distortion of MR images can adversely affect the accuracy of , the three-point-Dixon MR pulse sequence is evaluated as a mean of estimating the magnetic field inhomogeneity, and hence potential geometrical errors in images.
Finally, an experimental comparison of mechanical and optical localizing devices is described, resulting in a quantitative estimate of the precision and accuracy of both systems.