Technical reports

Automatic signal classification in fluorescence in-situ hybridization images

Boaz Lerner, William Clocksin, Seema Dhanjal, Maj Hultén, Christipher Bishop

May 1999, 24 pages

DOI: 10.48456/tr-466

Abstract

Previous systems for dot counting in fluorescence in-situ hybridization (FISH) images have relied on an automatic focusing method for obtaining a clearly defined image. Because signals are distributed in three dimensions within the nucleus and artifacts such as debris and background fluorescence can attract the focusing method , valid signals can be left unfocused or unseen. This leads to dot counting errors, which increase with the number of probes. The approach described here dispenses with automatic focusing, and instead relies on a larger statistical sample of the specimen at a fixed focal plane. Images across the specimen can be obtained in significantly less time if a fixed focal plane is used. A trainable classifier based on a neural network is used to discriminate between valid and artifact signals represented by a set of features. This improves on previous classification schemes that are based on non-adaptable decision boundaries and are trained using only examples of valid signals. Trained by examples of valid and artifact signals, three NN classifiers, two of them hierarchical, each achieve between 83% and 87% classification accuracy on unseen data. When data is pre-discriminated in this way, errors in dot counting can be significantly reduced.

Full text

PDF (1.2 MB)

BibTeX record

@TechReport{UCAM-CL-TR-466,
  author =	 {Lerner, Boaz and Clocksin, William and Dhanjal, Seema and
          	  Hult{\'e}n, Maj and Bishop, Christipher},
  title = 	 {{Automatic signal classification in fluorescence in-situ
         	   hybridization images}},
  year = 	 1999,
  month = 	 may,
  url = 	 {https://www.cl.cam.ac.uk/techreports/UCAM-CL-TR-466.pdf},
  institution =  {University of Cambridge, Computer Laboratory},
  doi = 	 {10.48456/tr-466},
  number = 	 {UCAM-CL-TR-466}
}