Automated Chromosome Classification for Karyotype Construction Using Deep Learning

• Laura G. Speggiorin UFRGS-HCPA
• Thayne W. Kowalski UFRGS-HCPA
• Claudio R. Jung UFRGS
• Mariana Recamonde-Mendoza UFRGS-HCPA

DOI: https://doi.org/10.5753/eramiars.2025.16745

Resumo

This project aims to automate human karyotype generation to optimize cytogenetic analysis, seeking to improve accuracy and reduce manual workload. Using a public dataset, the YOLOv10 model was employed for chromosome detection, demonstrating a mean average performance of 99.4%. For segmentation, the SAM model was used as a proof of concept, proving very successful in most cases for generating masks, although it faces challenges with overlapping chromosomes. In conclusion, the detection achieved excellent performance, and segmentation is promising with room for improvement.

Referências

Erwinsyah, R., Riandi, and Nurjhani, M. (2017). Relevance of human chromosome analysis activities against mutation concept in genetics course. IOP Conf. Ser. Mater. Sci. Eng., 180:012285.

Kirillov, A., Mintun, E., Ravi, N., Mao, H., Rolland, C., Gustafson, L., Xiao, T., Whitehead, S., Berg, A. C., Lo, W.-Y., Dollár, P., and Girshick, R. (2023). Segment anything.

Remani Sathyan, R., Chandrasekhara Menon, G., S, H., Thampi, R., and Duraisamy, J. H. (2022). Traditional and deep-based techniques for end-to-end automated karyotyping: A review. Expert Systems, 39(3):e12799.

Tseng, C.-H. L.-E. K.-J. (2022). CIL:54816, homo sapiens linnaeus, 1758, epithelial cell.

Wang, A., Chen, H., Liu, L., Chen, K., Lin, Z., Han, J., and Ding, G. (2024). YOLOv10: Real-time end-to-end object detection.