There are 46 chromosomes in a normal human cell. Each chromosome is composed of a long, single strand of DNA. At the right stage of cell division, the chromosomes condense and are observable by light microscopy. Certain genetic abnormalities can be observed in this manner and used to diagnose rare genetic diseases such as Down syndrome. It is often the case that the pattern of human chromosomes (that is, the karyotype) is compiled by hand.
There is some interest in automating this process using computational algorithms. Previous work in this area has focused almost exclusively on neural network approaches. Biyani and others argue that neural networks are suboptimal for this problem. This paper presents a maximum likelihood approach to the chromosome classification problem. The authors propose a three-dimensional assignment approach that uses a Lagrangian-type relaxation method for optimization. The authors were able to show that their method performed better than other methods in this domain.
It would be nice to eventually see this algorithm included in an open source software package that could be routinely used in cytogenetic laboratories that generate karyotypes from human cells for the purpose of genetic analysis and disease diagnosis. Integration of this algorithm into clinical practice will be the ultimate test of its validity and usefulness.