Feature normalization via expectation maximization and unsupervised nonparametric classification for M-FISH...

Area of Science:

• Genetics
• Computational Biology
• Medical Imaging

Background:

• Multicolor fluorescence in situ hybridization (M-FISH) enables color karyotyping for analyzing human chromosome abnormalities.
• Accurate classification of chromosome pixels is crucial for M-FISH success.
• Variations in intensity distributions between M-FISH images lead to misclassifications.

Purpose of the Study:

• To introduce a novel feature normalization method for M-FISH images.
• To develop an unsupervised, nonparametric classification method for M-FISH data.
• To improve pixel classification accuracy in M-FISH analysis.

Main Methods:

• Implemented an Expectation-Maximization (EM) algorithm for feature normalization.
• Developed an unsupervised, nonparametric classification approach for M-FISH images.
• Evaluated classification accuracy before and after EM normalization.

Main Results:

• The EM normalization method effectively reduces feature distribution differences among M-FISH images.
• The new unsupervised classifier achieves accuracy comparable to maximum-likelihood methods.
• Pixel classification accuracy improved by 20% after EM normalization.

Conclusions:

• The proposed EM normalization significantly enhances M-FISH pixel classification accuracy.
• The unsupervised classification method is convenient, especially when ground truth data is unavailable.
• This approach offers a robust solution for improving M-FISH data analysis and interpretation.