Efficient karyotyping of metaphase chromosomes using incremental learning

Automated karyotyping for chromosome classification is an essential task in cytogenetics for diagnosis of genetic disorders and has therefore been an important pattern recognition problem. The existing learning approaches generally discard the previously acquired knowledge and often require retraining, leading to space and time complexities. Incremental learning methods have gained popularity in the current learning scenarios to deal with these issues. This study proposes a novel approach of incremental learning for chromosomes classification for automated karyotyping of metaphase chromosomes. It addresses the issue of catastrophic forgetting with the generation of new class and performs knowledge amassing to classify the chromosomes in Denver groups (A-G). The adaptive nature of the proposed method contributes to its sustained accuracy even for dynamically changing data. An average classification accuracy of 97% is achieved with experimentation on 1800 chromosomes from a publicly available database.