Automated System for Chromosome Karyotyping Detection

Abstract

In this research, we present a system aimed at identifying and categorizing chromosomes within the context of karyotyping. Leveraging the YOLOv8(You Only Look Once) object detection framework, our approach focuses on training the system to recognize and classify individual chromosomes through exposure to numerous images containing these genetic structures. The developed system offers fast operation, reducing the time required for chromosome analysis, and high accuracy, minimizing inherent errors in manual analysis. This reduces the result delivery time in clinical applications. In the development process, we made use of a dataset comprising annotated chromosome images as the training material for our YOLO model. Through fine-tuning, we achieved a Mean Average Precision(mAP) of 80.7% for an Intersection over Union(IoU) threshold of 0.5 suggesting remarkable precision and recall rates, minimizing misclassifications while consuming less amount of resources, ensuring reliable chromosome detection and classification that can be established even on resource-constrained devices. The output obtained from the automated system can be used to confirm chromosomal abnormalities in the input samples. The potential applications of our system empower medical professionals with an improved understanding of genetic conditions thereby resulting in a more accurate diagnosis and accelerating genetic research.