IHC Color Histograms for Unsupervised Ki67 Proliferation Index Calculation

Area of Science:

• Digital Pathology
• Computational Biology
• Oncology

Background:

• Automated image analysis for Ki67 breast cancer pathology is valuable but challenged by multicenter data variability.
• Differences in staining, digitization, and preparation affect image quality and color consistency.

Purpose of the Study:

• To develop a robust, unsupervised color separation framework for analyzing Ki67 and hematoxylin in multicenter digital pathology images.
• To improve the accuracy and reliability of proliferation index quantification in breast cancer diagnostics.

Main Methods:

• Proposed a novel unsupervised color separation framework utilizing the IHC color histogram (IHCCH).
• Implemented an "overstaining" threshold and an automated nuclei radius estimator for enhanced detection and adjustment.
• Validated the method against manually labeled ground truth data and diverse datasets from the Protein Atlas.

Main Results:

• Achieved an average proliferation index difference of 3.25% compared to ground truth.
• Demonstrated 92.5% accuracy in classifying TMA images against PI ranges (74 out of 80).
• Outperformed color-deconvolution and machine learning methods in consistency and PI quantification accuracy.

Conclusions:

• The proposed IHCCH framework offers a robust and accurate solution for Ki67 analysis in multicenter breast cancer digital pathology.
• The method enhances diagnostic efficiency and reliability, addressing key challenges in real-world data variability.
• This approach shows significant potential for integration into routine diagnostic workflows, improving pathologist workload and diagnostic precision.