Technical reports

A Bayesian methodology and probability density estimation for fluorescence in-situ hybridization signal classification

Boaz Lerner

October 1999, 31 pages

DOI: 10.48456/tr-474

Abstract

Previous research has indicated the significance of accurate classification of flourescence in-situ hybridization (FISH) signals when images are captured in a fixed focal plane without relying on an auto-focusing mechanism. Based on well-discriminating features and a trainable neural network (NN) classifier, a previous system enabled highly-accurate classification of valid signals and artifacts of two fluorophores. However, since training and optimisation of an NN require extensive resources and experimentation, we investigate in this work a simpler alternative for the NN classifier – the naive Bayesian classifier (NBC). The Bayesian methodology together with an independence assumption allow the NBC to predict the a posteriori probability of class membership using estimated class-conditional densities. Densities measured by three methods: single Gaussian estimation (SGE; parametric method), Gaussian mixture model (GMM; semi-parametic method) and kernel density estimation (KDE; non-parametric method) are evaluated for this purpose. The accuracy of the NBC employing data modelled by SGE is found to be similar to that based on GMM, slightly inferior to that based on KDE but widely inferior to that of the NN. Therefore, when supporting the two classifiers, the system enables a trade-off between the NN performance and the NBC simplicity. Finally, the evaluation of the NBC accuracy provides a mechanism for both model and feature selection.

Full text

PDF (1.3 MB)

BibTeX record

@TechReport{UCAM-CL-TR-474,
  author =	 {Lerner, Boaz},
  title = 	 {{A Bayesian methodology and probability density estimation
         	   for fluorescence in-situ hybridization signal
         	   classification}},
  year = 	 1999,
  month = 	 oct,
  url = 	 {https://www.cl.cam.ac.uk/techreports/UCAM-CL-TR-474.pdf},
  institution =  {University of Cambridge, Computer Laboratory},
  doi = 	 {10.48456/tr-474},
  number = 	 {UCAM-CL-TR-474}
}