Radiologic localization of pathological lesions using image registration

The ability to accurately localize pathologic lesions has immediate practical benefits from both a therapeutic and diagnostic perspective. Pathologic lesions are often poorly visualized in the surgical field, especially at the margins, or else they have shifted in space from their pre-operative diagnostic imaging location, making accurate visual localization of lesions difficult. In addition, diagnostic imaging modalities have different strengths and weaknesses necessitating the need for multiple examinations that are often not easy to align. The methods currently used to overcome these difficulties such as direct lesion stimulation, biopsy or intraoperative imaging are crude, time-consuming or not widely available.;A novel image registration algorithm that can be used to localize radiologic lesions both accurately and rapidly was developed. The algorithm used an Expectation-Maximization (EM) approach to find the optimal matching matrix and registration transformation for image alignment between two three-dimensional data sets, although the algorithm is arbitrary enough to work with any n-dimensional source and target space. The two data sets that were aligned were the point data extracted from Computed Tomography (CT) or Magnetic Resonance Imaging (MRI) diagnostic volumes and optional point data from a tracking device (such as an optical range tracker). The data-sets can be both pre-operative and intraoperative allowing for therapeutic as well as diagnostic applications.;Two practical applications of this algorithm were tested. The first application involved the alignment of CT and MRI scans in prostate brachytherapy patients. The algorithm was able to accurately align the CT-visible brachytherapy seeds with the corresponding voids visible in the MRI. This alignment provided a statistically significant improvement over the Iterative Closest Points (ICP) registration algorithm. The second application was the alignment of Electroencephalogram (EEG) electrodes to MRI scans in a pediatric patient model. The use of EEG electrodes solely to align to the MRI surface did not provide a statistically significant improvement. However, the addition of fiducial markers allowed for the proper alignment of the data sets.