Low SNR Computational Pattern Detection Applied to Multi-Spectral 3D Molecular Dynamics
MetadataShow full item record
CitationChang, Frederick. 2018. Low SNR Computational Pattern Detection Applied to Multi-Spectral 3D Molecular Dynamics. Doctoral dissertation, Harvard University, Graduate School of Arts & Sciences.
AbstractWe describe an algorithm for computational pattern detection with expanded performance into low Signal-to-Noise Ratio (SNR) regimes. Our approach consists of a Two-Stage Likelihood Pipeline (TSLP) and is applicable to large N-dimensional datasets. We apply this approach to the detection and localization of 3D fluorescent point sources (“spots”) in datasets generated by fluorescent microscopy. We demonstrate by in silico benchmarking that our approach is both maximal in sensitivity and selectivity in detection and minimal in error in photometry and localization of spots. Most importantly, the probability distribution of the Likelihood ratio is empirically derived, therefore the detection of spots in varying background conditions consists of one parameter. Since imaging in low SNR regimes corresponds to imaging with low excitation energies, our approach can enable long timescale imaging of 3D fluorescent spots at high temporal resolution. We illustrate this capability by analyzing 3D in vivo dynamics of fluorescently tagged single molecules and oligomeric complexes. Additionally, imaging in 3D and/or multiple colors (or ND) is achieved with a minimal increase in excitation energy as compared to 2D single color imaging.
Citable link to this pagehttp://nrs.harvard.edu/urn-3:HUL.InstRepos:42015127
- FAS Theses and Dissertations