Robust Analysis of Spot Array Images (bibtex)
by Norbert Braendle
Abstract:
Computer-aided image analysis deals with the automatic recovery of visual information from digital images. Recent image analysis research is trying to find more general and more efficient algorithms. Given a digital image with a certain contents or certain problem domain, it is still mandatory to manually evaluate different approaches and processing sequences to extract useful and plausible information from the image. Once a generic image analysis approach for the problem domain has been found, the method can often be applied independent of the operator's intuition and previous experience. Optical character recognition (OCR) is an example for a successful development from initial research to off-the-shelf products. This work provides a generic image analysis system for a class of images having a characteristic contents denoted as \em spot arrays. Spots are defined as simply connected, irregularly shaped regions lighter (or darker) than their background. Representatives of this image class include images of Braille paper for blind persons and DNA arrays - a tool of modern biotechnology. Analysis of spot array image has three main tasks. The first task is to detect the spots present in the image and therefore deals with the spatial localization process. The spots in the image are located on a grid which may be distorted in the course of the image production process. The second task therefore deals with the fitting of a grid to the detected spots, such that they can be correctly addressed. Once the spots are detected and addressed, they are characterized by their shape, intensity and local background. The automatic image analysis presented in this work is composed of a set of tools arranged in a general framework. This general framework enables to analyze spot arrays of high spot density with possibly multiple overlapping spots. Furthermore, the concept is robust in order to cope with outliers in the spot array and artifacts like image contaminations. These requirements can be fulfilled by robust statistical models: A key principle of grid fitting is to fit straight line models to the rows and columns of the grid. The input for the straight line models is given by a maximum search in matched filter response. Spot characterization is performed by fitting a parametric spot model to the corresponding pixels with the help of a robust M-estimator. In a consecutive step, a semi-parametric fit is possible in order to cope with deviations from the spot model assumptions. Analysis of DNA array images serves as a demonstration of the presented general framework. Here, the intensity of a spot represents the amount of genetic material bound to the corresponding array element. The ultimate image analysis goal of this application is to quantify as exactly as possible the intensity of tens of thousands of possibly overlapping spots. The output of DNA array image analysis yields the raw data for the discovery of specific genes and the genetic control system of organisms. The results of DNA array images demonstrate the successful application of the framework presented in this work on thousands of images resulting from various biological experiments.
Reference:
Robust Analysis of Spot Array Images (Norbert Braendle), Technical report, PRIP, TU Wien, 2002.
Bibtex Entry:
@TechReport{PTR-Braendle02a,
  author =	 "Norbert Braendle",
  institution =	 "PRIP, TU Wien",
  number =	 "PRIP-TR-070",
  title =	 "Robust {A}nalysis of {S}pot {A}rray {I}mages",
  year =	 "2002",
  url =		 "ftp://ftp.prip.tuwien.ac.at/pub/publications/trs/tr70.pdf",
  abstract =	 " Computer-aided image analysis deals with the
                  automatic recovery of visual information from
                  digital images. Recent image analysis research is
                  trying to find more general and more efficient
                  algorithms. Given a digital image with a certain
                  contents or certain problem domain, it is still
                  mandatory to manually evaluate different approaches
                  and processing sequences to extract useful and
                  plausible information from the image. Once a generic
                  image analysis approach for the problem domain has
                  been found, the method can often be applied
                  independent of the operator's intuition and previous
                  experience. Optical character recognition (OCR) is
                  an example for a successful development from initial
                  research to off-the-shelf products. This work
                  provides a generic image analysis system for a class
                  of images having a characteristic contents denoted
                  as {\em spot arrays}. Spots are defined as simply
                  connected, irregularly shaped regions lighter (or
                  darker) than their background. Representatives of
                  this image class include images of Braille paper for
                  blind persons and DNA arrays - a tool of modern
                  biotechnology. Analysis of spot array image has
                  three main tasks. The first task is to detect the
                  spots present in the image and therefore deals with
                  the spatial localization process. The spots in the
                  image are located on a grid which may be distorted
                  in the course of the image production process. The
                  second task therefore deals with the fitting of a
                  grid to the detected spots, such that they can be
                  correctly addressed. Once the spots are detected and
                  addressed, they are characterized by their shape,
                  intensity and local background. The automatic image
                  analysis presented in this work is composed of a set
                  of tools arranged in a general framework. This
                  general framework enables to analyze spot arrays of
                  high spot density with possibly multiple overlapping
                  spots. Furthermore, the concept is robust in order
                  to cope with outliers in the spot array and
                  artifacts like image contaminations. These
                  requirements can be fulfilled by robust statistical
                  models: A key principle of grid fitting is to fit
                  straight line models to the rows and columns of the
                  grid. The input for the straight line models is
                  given by a maximum search in matched filter
                  response. Spot characterization is performed by
                  fitting a parametric spot model to the corresponding
                  pixels with the help of a robust M-estimator. In a
                  consecutive step, a semi-parametric fit is possible
                  in order to cope with deviations from the spot model
                  assumptions. Analysis of DNA array images serves as
                  a demonstration of the presented general
                  framework. Here, the intensity of a spot represents
                  the amount of genetic material bound to the
                  corresponding array element. The ultimate image
                  analysis goal of this application is to quantify as
                  exactly as possible the intensity of tens of
                  thousands of possibly overlapping spots. The output
                  of DNA array image analysis yields the raw data for
                  the discovery of specific genes and the genetic
                  control system of organisms. The results of DNA
                  array images demonstrate the successful application
                  of the framework presented in this work on thousands
                  of images resulting from various biological
                  experiments. ",
}
Powered by bibtexbrowser