Automated Detection of Dicentric Chromosomes Using Deep Learning: A Scalable Solution for Cytogenetic Analysis

The detection of dicentric chromosomes in metaphase images is highly essential for radiation biodosimetry and genetic studies. Conventional approaches of manual identification is labor-intensive and prone to errors, necessitating an automated and robust detection system. In this study, we propose a hybrid deep learning framework that combines a Two-Way Convolutional Neural Network (CNN) for feature extraction with a Faster Region-Based Convolutional Neural Network (Faster RCNN) for object detection. The proposed Two-Way CNN architecture extracts both local structural details and global morphological features ensuring high-quality feature representation. These enhanced feature maps are fed into the Faster R-CNN, which detects dicentric chromosomes by localizing bounding boxes around them. The model is trained on metaphase images annotated with dicentric and normal chromosomes. Experimental results demonstrate that our approach achieves high detection accuracy and robustness, outperforming traditional image processing techniques and standalone CNN-based classifiers. The integration of Two-Way CNN with Faster RCNN enhances both classification and localization, making it a promising tool for automated chromosome analysis in medical and biological research.