Exploring HSI imagery for overlapped chromosomes segmentation.

Area of Science:

• Cytogenetics
• Biomedical Imaging
• Computational Biology

Background:

• Accurate segmentation of overlapping chromosomes in metaphase images is crucial for cytogenetic analysis.
• Conventional RGB imaging faces limitations due to insufficient spectral contrast, hindering segmentation accuracy.
• Hyperspectral imaging (HSI) offers a potential solution by leveraging detailed spectral information.

Purpose of the Study:

• To evaluate the effectiveness of hyperspectral imaging (HSI) in improving the segmentation accuracy of overlapping chromosomes.
• To compare the performance of segmentation models trained on HSI data versus traditional RGB data.
• To introduce a new annotated dataset for hyperspectral chromosome segmentation research.

Main Methods:

• Acquisition and annotation of a dedicated dataset of overlapping chromosomes using a hyperspectral microscopy system.
• Development and training of multiple baseline segmentation models.
• Benchmarking model performance using both RGB and HSI data, employing metrics such as the Dice score.

Main Results:

• HSI-based segmentation models consistently outperformed RGB-based models.
• The highest Dice score achieved by HSI models reached 92%, demonstrating superior segmentation accuracy.
• The study validated the benefit of spectral data for chromosome segmentation.

Conclusions:

• Hyperspectral imaging (HSI) shows significant potential for enhancing cytogenetic image analysis.
• HSI provides a valuable alternative to conventional imaging for accurate chromosome segmentation.
• The developed benchmark dataset will facilitate future research in spectral segmentation of cytogenetic samples.