Comparative analysis of Ki-67 labeling index morphometry using deep learning, conventional image analysis, and manual...

Area of Science:

• Oncology
• Pathology
• Medical Imaging

Background:

• The Ki-67 labeling index is crucial for cancer prognosis and diagnosis.
• Current manual and digital image analysis (DIA) methods for Ki-67 estimation have limitations.
• Accurate Ki-67 quantification is essential for reliable clinical decision-making.

Purpose of the Study:

• To evaluate and compare the performance of different DIA systems for Ki-67 counting.
• To assess the impact of various annotation methods on Ki-67 positivity estimation.
• To compare DIA system results against conventional manual counting by pathologists.

Main Methods:

• 239 stomach cancer tissue microarray (TMA) cores were stained for Ki-67 and scanned.
• Three annotation methods were used: whole TMA core, box selection, and hand-free epithelium selection.
• Ki-67 counting was performed using DIA systems (3DHistech, Roche, aetherAI) and manual pathologist counts.

Main Results:

• Annotation methods yielded lower Ki-67 positivity compared to manual pathologist counts.
• Roche system is preferred for whole TMA analysis; aetherAI excels over box selection.
• 3DHistech demonstrates the highest accuracy for hands-free epithelium selection.
• Manual counts showed high inter-pathologist agreement (ICC=0.93).

Conclusions:

• The choice of annotation method significantly impacts Ki-67 positivity determination.
• Different DIA systems and annotation strategies have distinct strengths and weaknesses.
• Laboratories should validate multiple DIA approaches before routine implementation for Ki-67 assessment.