Advanced statistical matrices for texture characterization: application to cell classification.

Area of Science:

• Quantitative cell biology
• Image analysis
• Biomedical imaging

Background:

• Texture analysis is crucial for quantitative cell biology.
• Existing gray level size zone matrix (SZM) methods have limitations in information processing and parameter reduction.
• Accurate texture classification is needed for analyzing cellular structures like DNA organization during mitosis.

Purpose of the Study:

• To develop novel variants of the gray level size zone matrix (SZM) texture descriptor.
• To improve texture classification accuracy for supervised analysis of cell images.
• To adapt SZM for characterizing DNA organization and fibrous textures.

Main Methods:

• Developed improved SZM variants by merging gray-level quantizations to enhance information processing and reduce parameters.
• Designed new descriptors specifically for supervised cell texture classification tasks.
• Applied generalized matrix structures for fibrous texture analysis, using length/orientation pairs instead of intensity/size.

Main Results:

• The enhanced SZM descriptors demonstrated improved performance in supervised cell texture classification.
• The new methods were validated using two distinct quantitative cell biology databases.
• Characterization of DNA organization during mitosis was achieved through radial distribution analysis of zone intensities.

Conclusions:

• The proposed SZM variants offer a more informative and parameter-efficient approach to texture analysis in cell biology.
• These descriptors are effective for supervised classification and characterizing complex cellular structures.
• The generalized matrix structure opens new avenues for fibrous texture analysis.