Related Experiment Video
Updated: May 20, 2026
08:48
High-throughput Physical Mapping of Chromosomes using Automated in situ Hybridization
Published on: June 28, 2012
Automated karyotyping and structural anomaly detection through a hybrid multi-stage deep learning framework
Carolina Rosas-Alatriste1, Noé Oswaldo Rodríguez-Rodríguez1, Amadeo José Argüelles-Cruz1
1Instituto Politécnico Nacional (IPN), Centro de Investigación en Computación (CIC), Ciudad de México, 07738, México.
Scientific Reports
|May 18, 2026
Summary
This study introduces a deep learning framework for semi-automated karyotype generation, improving chromosome analysis accuracy and efficiency. The system effectively screens for numerical and structural chromosomal anomalies, reducing manual workload in cytogenetics.
Area of Science:
- Genetics
- Computational Biology
- Medical Imaging
Background:
- Manual karyotype analysis is time-consuming and requires specialized expertise.
- Existing machine learning methods often focus on isolated steps, not the complete cytogenetic workflow.
- A clinically aligned, semi-automated solution is needed to improve efficiency and accuracy.
Purpose of the Study:
- To develop a stability-aware, multi-stage deep learning framework for semi-automated karyotype generation.
- To integrate chromosome detection, homologous-pair assignment, karyotype assembly, and anomaly screening into a single pipeline.
- To enable efficient screening of numerical and structural chromosomal anomalies.
Main Methods:
- A multi-stage deep learning pipeline incorporating quality control, YOLOv8 for detection and classification, and a convolutional autoencoder for anomaly screening.
- Integration of YOLOv8 and ResNet-50 for homologous-pair assignment.
- Unsupervised structural anomaly screening using a convolutional autoencoder trained on normal chromosomes.
Main Results:
- Achieved 98.03% detection accuracy and 94.01% homologous-pair classification accuracy.
- End-to-end numerical accuracy reached up to 94.01%.
- The autoencoder effectively separated normal and structurally anomalous reconstruction errors, enabling unsupervised screening.
Conclusions:
- The proposed framework offers a clinically interpretable and operationally coherent system for semi-automated karyotype generation.
- Stability-aware pipeline design supports both numerical and structural anomaly detection.
- The system reduces manual workload while preserving diagnostic transparency in cytogenetics.
