Robust Papanicolaou Stain Quantification Insensitive to Imaging System Variations by Sparsity-based Stain Unmixing.

Area of Science:

• Digital Pathology
• Computational Cytology
• Biomedical Imaging

Background:

• Papanicolaou (Pap) stain is crucial for cytological examinations.
• Digital cytology aims for objective stain interpretation via quantitative measurements.
• Previous methods unmixed Pap stain in RGB images but required validation on real systems.

Purpose of the Study:

• To validate the robustness of Pap stain unmixing for quantitative stain analysis in actual RGB imaging systems.
• To assess the method's performance across different imaging systems with inherent color variations.
• To apply the validated technique for classifying specific cell types, aiding diagnosis.

Main Methods:

• Developed a sparsity-based regularization method for Pap stain unmixing in RGB images.
• Extracted stain matrices from single stain images and normalized using robust maximum values for color variation resilience.
• Trained a classifier using RGB images derived from multispectral data for cell classification.

Main Results:

• Demonstrated robust stain quantification across different imaging systems, overcoming color variations.
• Achieved high accuracy in classifying cytoplasmic mucin in lobular endocervical glandular hyperplasia and normal endocervical cells.
• Successfully applied the technique to whole slide images from a practical scanner.

Conclusions:

• Pap stain unmixing is effective and robust for quantitative analysis in standard RGB digital cytology.
• The validated method enables objective assessment and accurate classification of cytological samples.
• This technique holds promise for improving diagnostic criteria and objectivity in pathology.