Automated recognition of irradiated chromosome images

The problem of monitoring health related radiation damage is an important one for future and a system to detect chromosomal aberrations associated with radiation is extremely useful for monitoring populations at risk. In general, the most useful indicator of radiation damage at cell level is the occurrence of ring, dicentric, fragment, and ring fragment chromosomes. The skeleton transformation is used for describing shapes of these chromosomes. A ''cutoff Level Independent Skeletonization'' technique is proposed, discussed, and compared against Hilditch's and Montanari's techniques for obtaining a skeleton from a digitized grey level picture. An automated system using this approach for recognizing irradiated chromosomes is described in this dissertation. The test results of the system on 25 irradiated cells are included. The system requires 71.8 sec of CP time (CDC 6500) per cell and achieve s about 94 percent accuracy of correct classification of chromosomes. A ''concave-segment'' method to detect touching and overlapping chromosomes is also proposed and discussed. The test results of this method on 10 cells are included. It achieves 96 percent accuracy in detecting such cases. A method is also outlined to separate touching chromosomes and a special case of overlapping chromosomes