High quality three-dimensional bioluminescence tomographic (BLT) images, if available, would constitute a major advance and provide much more useful information than the two-dimensional bioluminescence images that are frequently used today. To-date, high quality BLT images have not been available, largely because of the poor quality of the data being input into the reconstruction process. Many significant confounds are not routinely corrected for and the noise in this data is unnecessarily large and poorly distributed. Moreover, many of the design choices affecting image quality are not well considered, including choices regarding the number and type of filters used when making multispectral measurements and choices regarding the frequency and uniformity of the sampling of both the range and domain of the BLT inverse problem. Finally, progress in BLT image quality is difficult to gauge owing to a lack of realistic gold-standard references that engage the full complexity and uncertainty within a small animal BLT imaging experiment.
Within this dissertation, I address all of these issues. I develop a Cerenkov-based gold-standard wherein a Positron Emission Tomography (PET) image can be used to gauge improvements in the accuracy of BLT reconstruction algorithms. In the process of creating this reference, I discover and describe corrections for several confounds that if left uncorrected would introduce artifacts into the BLT images. This includes corrections for the angle of the animal’s skin surface relative to the camera, for the height of each point on the skin surface relative to the focal plane, and for the variation in bioluminescence intensity as a function of luciferin concentration over time. Once applied, I go on to derive equations and algorithms that when employed are able to minimize the noise in the final images under the constraints of a multispectral BLT data acquisition. These equations and algorithms allow for an optimal choice of filters to be made and for the acquisition time to be optimally distributed among those filtered measurements. These optimizations make use of Barrett’s and Moore-Penrose pseudoinverse matrices which also come into play in a paradigm I describe that can be used to guide choices regarding sampling of the domain and range.
| Identifer | oai:union.ndltd.org:columbia.edu/oai:academiccommons.columbia.edu:10.7916/D8001DMF |
| Date | January 2018 |
| Creators | Beattie, Bradley |
| Source Sets | Columbia University |
| Language | English |
| Detected Language | English |
| Type | Theses |
Page generated in 0.0019 seconds