BR-ChromNet: Banding resolution localization of chromosome structural abnormality with conditional random field

Area of Science:

• Medical Genetics
• Computational Biology
• Bioinformatics

Background:

• Accurate detection of chromosome structural abnormalities is crucial for diagnosing genetic disorders.
• Current computational methods often treat abnormality detection as binary classification, limiting precise localization.
• Existing approaches lack interpretability for cytogeneticists.

Purpose of the Study:

• To develop an innovative framework for detecting chromosome abnormalities with banding resolution.
• To precisely identify and mask specific abnormal chromosomal regions.
• To enhance the interpretability of prediction results for cytogeneticists.

Main Methods:

• A pixel-level abnormal mapping strategy guided by banding features.
• Integration of original image data and banding characteristics.
• An ensemble approach combining a discriminator and a conditional random field heatmap generator.

Main Results:

• Significantly reduced false positive rates in abnormality screening.
• Superior performance in sensitivity and segmentation accuracy compared to SOTA methods.
• Demonstrated cutting-edge effectiveness in identifying abnormal regions.

Conclusions:

• The BRChromNet framework offers significant clinical utility with high model interpretability.
• The model accurately identifies and masks abnormal chromosomal regions.
• Open-sourced availability facilitates further research and application.