Semantic segmentation of mFISH images using convolutional networks.

Area of Science:

• Genetics
• Computational Biology
• Medical Imaging

Background:

• Multicolor in situ hybridization (mFISH) is crucial for detecting chromosomal abnormalities.
• Manual interpretation of mFISH images is labor-intensive and prone to error.
• Previous machine learning attempts overlooked critical spatial information for chromosome identification.

Purpose of the Study:

• To develop an automated method for mFISH image interpretation using machine learning.
• To leverage both spatial and spectral information for accurate pixel-wise classification.
• To improve the efficiency and accuracy of genetic disease diagnosis through computer-aided analysis.

Main Methods:

• A fully convolutional semantic segmentation network was designed for end-to-end mFISH image analysis.
• The network processes images using both spatial and spectral data for pixel classification.
• Cross-validation was performed on a public dataset to evaluate the algorithm's performance.

Main Results:

• The semantic segmentation network achieved an average correct classification ratio (CCR) of 87.41%.
• This accuracy was obtained even without prior labeling of the test images.
• The performance surpasses previous state-of-the-art methods that required training on the same image data.

Conclusions:

• Fully convolutional semantic segmentation networks offer a promising approach for automated mFISH image analysis.
• This technology can significantly enhance computer-aided diagnosis of genetic diseases.
• The proposed method demonstrates superior performance compared to existing image analysis techniques.