ChromEDA: Chromosome classification by ensemble framework based domain adaptation.

Area of Science:

• Computational Biology and Bioinformatics
• Medical Imaging and Image Analysis
• Machine Learning and Artificial Intelligence

Background:

• Manual karyotype analysis for chromosomal abnormalities is labor-intensive and costly.
• Current deep learning models require large annotated datasets and struggle with domain shifts across different data sources.
• Domain shifts between datasets from various instruments and clinical agencies limit the use of public data resources.

Purpose of the Study:

• To develop an automated chromosome classification technology that addresses the limitations of manual analysis and current deep learning methods.
• To mitigate domain shift issues in chromosome image classification.
• To reduce the reliance on large amounts of annotated data for training.

Main Methods:

• Proposed ChromEDA, a novel ensemble learning framework.
• Incorporated soft pseudo-label learning, adversarial learning, and angle classification learning strategies.
• Conducted cross-domain classification experiments using both public and private datasets.

Main Results:

• ChromEDA effectively addresses the domain shift issue in chromosome classification.
• The framework outperforms existing methods in cross-domain chromosome classification tasks.
• Demonstrated the potential for unsupervised cross-domain algorithms in chromosome microscopic image processing.

Conclusions:

• ChromEDA offers a robust solution for automated chromosome classification, overcoming domain shift challenges.
• The method significantly reduces annotation requirements, facilitating faster and more efficient deployment.
• Banding pattern information is crucial for precise within-group chromosome classification.