An optimized image analysis algorithm for detecting nuclear signals in digital whole slides for histopathology

Area of Science:

• Digital pathology
• Biomarker quantification
• Computational pathology

Background:

• Nuclear biomarkers (ER, PR, Ki-67) are crucial for breast cancer prognosis and treatment selection.
• Current assessment relies on manual microscopic scoring, which can be subjective and inconsistent.
• Digital pathology offers potential for more objective and reproducible biomarker analysis.

Purpose of the Study:

• To develop and validate an image analysis algorithm for automated quantification of nuclear ER, PR, and Ki-67 in breast cancer.
• To assess the algorithm's reliability, accuracy, and processing speed compared to existing methods.
• To support pathologists in objective biomarker assessment, reducing inter- and intra-laboratory variability.

Main Methods:

• Development of a novel image analysis algorithm for whole slide quantification of nuclear immunostaining.
• Testing the algorithm on brightfield and fluorescent stained breast cancer samples.
• Validation of detection accuracy, precision, recall, and processing speed against manual scoring and open-source tools (QuPath, CellProfiler).

Main Results:

• The algorithm achieved high efficacy with a Precision Rate/Positive Predictive Value of 90.23% ± 4.29% and a Recall Rate/Sensitivity of 88.23% ± 4.84%.
• It demonstrated superior performance compared to QuPath and CellProfiler, with 6-7% higher Recall Rate and 4- to 30-fold faster processing speed.
• The method effectively separates overlapping signals and nuclei, and compensates for background noise.

Conclusions:

• The developed image analysis algorithm reliably quantifies nuclear biomarkers in digital breast cancer slides.
• This automated approach can significantly aid pathologists in biomarker assessment, minimizing subjective scoring biases.
• The system offers a robust and efficient tool for improving diagnostic consistency in breast cancer management.