Automatic Ki-67 counting using robust cell detection and online dictionary learning.

Area of Science:

• Oncology
• Computational Pathology
• Biomedical Imaging

Background:

• The Ki-67 proliferation index is crucial for assessing neuroendocrine tumor (NET) progression.
• Accurate Ki-67 assessment is challenging due to cellular heterogeneity in NETs.

Purpose of the Study:

• To develop an integrated, learning-based framework for accurate automatic Ki-67 counting in NETs.
• To improve the differentiation of tumor cells from non-tumor cells and quantify immunopositive tumor cells.

Main Methods:

• A robust cell counting and boundary delineation algorithm for localizing tumor and non-tumor cells.
• An online sparse dictionary learning method for selecting representative training samples.
• An automated framework for differentiating cell types and assessing Ki-67 index.

Main Results:

• The framework was validated on 46 NET cases, comparing automated counts with manual annotations by pathologists.
• The proposed method demonstrated high accuracy in automatic Ki-67 counting.
• Performance significantly surpassed existing automated methods.

Conclusions:

• The developed framework provides accurate and automated Ki-67 assessment for NETs.
• This approach offers a more reliable alternative to manual annotation for evaluating tumor cell proliferation.
• The method has the potential to enhance clinical decision-making in NET management.