Quantitative immunohistochemistry analysis of breast Ki67 based on artificial intelligence

Area of Science:

• Oncology
• Pathology
• Artificial Intelligence

Background:

• Breast cancer is a prevalent malignancy in women.
• Ki67 is a crucial biomarker for cell proliferation and malignancy assessment.
• Manual Ki67 evaluation is subjective, time-consuming, and labor-intensive.

Purpose of the Study:

• To develop and validate a deep learning-based quantitative analysis pipeline for Ki67 in breast cancer pathology.
• To overcome the limitations of manual assessment in accuracy and efficiency.

Main Methods:

• A deep learning approach was employed to analyze Ki67 pathological images.
• The pipeline involved tumor region identification, nucleus detection, and Ki67 positive/negative classification.
• Quantitative analysis calculated the proportion of positive cells.

Main Results:

• The system achieved high performance metrics: Dice coefficient of 0.848 for tumor segmentation, Average Precision of 0.817 for nucleus detection, and 96.66% accuracy for nucleus classification.
• Clinical validation demonstrated over tenfold improvement in diagnostic efficiency and high consistency (intra-group correlation coefficient: 0.964).

Conclusions:

• The developed deep learning system provides an accurate and efficient method for Ki67 quantitative analysis in breast cancer.
• This approach holds significant clinical value for improving breast cancer diagnosis and consistency.