KaryoXpert: An accurate chromosome segmentation and classification framework for karyotyping analysis without...

Area of Science:

• Cytogenetics
• Artificial Intelligence
• Computational Biology

Background:

• Automated karyotyping accelerates cytogenetic research by employing AI for chromosome segmentation and classification.
• Current methods require extensive instance-level data annotation and suffer from poor robustness during domain shifts.

Purpose of the Study:

• To develop an accurate and robust automated karyotyping framework, KaryoXpert.
• To overcome limitations of existing methods, specifically the need for manual annotations and susceptibility to domain shifts.

Main Methods:

• Proposed KaryoXpert, a segmentation framework combining morphology algorithms and deep learning for efficient training without manual mask annotations.
• Developed a classification model using metric learning to address inter-class similarity and batch effects.

Main Results:

• KaryoXpert achieved state-of-the-art performance in chromosome segmentation and classification.
• Demonstrated robustness against domain shifts and successful instance-level segmentation without annotated data.
• Successfully deployed for clinical karyotype diagnosis support.

Conclusions:

• KaryoXpert offers an efficient and robust solution for automated chromosome segmentation and classification in cytogenetics.
• The framework advances automated chromosome analysis and supports clinical diagnostic applications.