A novel metric for bone marrow cells chromosome pairing

Area of Science:

• Cytogenetics
• Medical image analysis
• Computational biology

Background:

• Automatic chromosome pairing is crucial for leukemia diagnostics but challenging due to distorted, low-quality bone marrow cell images.
• Traditional karyotyping methods struggle with image artifacts like blurring and undefined edges.

Purpose of the Study:

• To develop a new metric and an automatic pairing algorithm for bone marrow chromosomes to aid in leukemia diagnosis.
• To improve the discrimination power of G-banding patterns using mutual information.

Main Methods:

• A novel metric based on mutual information was proposed to enhance G-banding pattern dissimilarity.
• The chromosome pairing algorithm was formulated as a combinatorial optimization problem.
• Integer programming was used to solve the optimization task, minimizing homologous and maximizing nonhomologous chromosome distances.

Main Results:

• A new dataset, Lisbon-K1 (LK1), comprising 9200 low-quality bone marrow chromosomes, was created.
• The proposed algorithm demonstrated effectiveness in pairing chromosomes from challenging datasets, including LK1 and Grisan et al.
• Leave-one-out cross-validation was employed for rigorous testing and validation.

Conclusions:

• The developed automatic pairing algorithm shows promise for improving leukemia diagnostics.
• The novel mutual information-based metric enhances the classification of low-quality chromosome images.
• The study provides a robust method for handling difficult bone marrow chromosome image analysis.