Chromosome Karyotyping and Aneuploidy Detection using Deep Learning Networks

Abstract

Chromosome karyotype analysis is a vital role in the diagnosis of congenital disease, chromosomal aneuploidies. Karyotyping is defined as the process of categorizing chromosomes into one of 24 kinds. Chromosomal aneuploidies can be classified as structural and numerical. Numerical anomalies such as Down syndrome, Williams’s syndrome Pattu syndrome, Turner syndrome, klinefelter syndrome and certain kind of cancers detected with the help of karyotyping. It necessitates rigorous attention to detail and well-trained people. With these objectives in mind, in this work, we automated karyotyping and identified common numerical aneuploidies using deep learning on a dataset of 5474 distinct grayscale metaphase chromosomal impressions from the Bioimage chromosome categorization dataset (BioImLab), which is freely available online. Three stages are included in the proposed classification model. The first stage entails segmenting and individual chromosome identification using bounding box detection and modified binary mask methods. At this point, achieved individual chromosome detection accuracy is 99%. The second stage involves classifying chromosomes using a 144 deep and 1000 FC layer GLNet (GoogleNet) classifier, which has an accuracy rate of 96.33%. The last step is to accurately identify aneuploidies with a 98% detection rate. A robust automatic classification system has been developed. Many scholars have proposed various karyotyping approaches. This research compares and contrasts the outcomes of several approaches with greater accuracy.