Detection of lobular structures in normal breast tissue.

Area of Science:

• Computational pathology
• Digital pathology
• Biomedical image analysis

Background:

• Evaluating immune cells in histological sections is crucial for inflammatory condition research.
• Objective and consistent quantitative studies require automated detection of specific regions of interest (ROIs).
• This study focuses on automating the detection of lobular structures in human normal breast tissue whole slide images (WSIs).

Purpose of the Study:

• To compare different automated image analysis methods for detecting lobular structures in WSIs.
• To assess the feasibility and performance of various approaches for ROI detection in breast tissue.
• To provide insights for selecting appropriate automated ROI detection methods for biomarker research.

Main Methods:

• Three automated image analysis methods were evaluated on normal breast tissue from nine healthy patients.
• Methods included a bottom-up cell-based approach, a top-down texture classification approach, and deep learning-based texture classification.
• Immunohistochemically stained WSIs were used to detect lobular structures and analyze cell densities within ROIs.

Main Results:

• All three evaluated methods achieved comparable quality in automated lobular structure detection.
• Deep learning showed a minor advantage in sensitivity, texture classification in specificity, and the bottom-up approach in processing time.
• Combining the outputs from different approaches further improved the precision of lobular structure detection.

Conclusions:

• Automated detection of ROIs, specifically lobular structures, is feasible using various image analysis techniques.
• The choice of method should align with specific biomarker research needs regarding sensitivity, specificity, or speed.
• Detected ROIs can serve as a foundation for quantifying immune cell infiltration within lobular structures.