Computer-assisted detection of chromosomal aberrations for the purpose of establishing a biological dosimeter

Using a special high-resolution microscopy apparatus, digital light microscope images of human chromosomes were generated, and specific image processing algorithms were developed. The pattern recognition process for computer-assisted detection of specific, radiation-induced chromosomal aberrations comprises three steps: First, the orientation of the segmented objects is defined and corrected. For date reduction purposes, the individual chromosomes are reduced to a few basic types containing typical information. After a linear transformation step, the characteristic parameters thus derived form a parameter vector for statistical classification. The method was well suited for distinguishing normal chromosomes from chromosomal aberrations. 94% of the objects were identified correctly. To achieve even higher accuracy, quality standards were set by which suspectedly misclassified objects can be re-investigated in dialog by the human observer. Implementation of the program system for parameter extraction on a fast polyprocessor system opens up a realistic chance of reducing the computing time for dose estimates to about one hour. (orig./MG)