Deep Learning-Based Pathology Image Analysis Enhances Magee Feature Correlation With Oncotype DX Breast Recurrence...

Area of Science:

• Digital Pathology
• Machine Learning in Oncology
• Breast Cancer Biomarkers

Background:

• The Oncotype DX Recurrence Score (RS) is crucial for predicting chemotherapy benefit in ER-positive breast cancer.
• Existing Magee equations correlate with RS, but their predictive power can be improved.
• Deep learning (DL) offers a novel approach to analyze histopathology images for enhanced RS prediction.

Purpose of the Study:

• To investigate if DL-based analysis of histology images can improve the correlation with Oncotype DX Recurrence Score (RS).
• To compare the predictive performance of DL-derived image features against traditional Magee features.
• To determine if combining DL features with Magee features enhances RS prediction.

Main Methods:

• Retrieved 382 ER-positive breast cancer cases with RS data.
• Developed DL models to identify and segment tumor cells and tumor-infiltrating lymphocytes (TILs) in H&E stained whole slide images (WSIs).
• Extracted image features (e.g., cell counts, nuclear grades) and compared prediction models using Magee features alone versus combined Magee and DL-derived features.

Main Results:

• DL-based analysis showed significant correlations with actual RS (Pearson's r = 0.7058 and 0.5041 on validation sets).
• Adjusted R-squared values for RS prediction were enhanced from 0.3442/0.2167 (Magee features alone) to 0.4431/0.2182 when WSI-derived DL features were included.
• These findings indicate improved predictive accuracy with the integration of digital pathology features.

Conclusions:

• DL-based digital pathology features significantly enhance the correlation with Oncotype DX Recurrence Score (RS).
• Integrating DL-derived image features with traditional Magee features improves the prediction of RS in ER-positive breast cancer.
• This approach holds promise for refining treatment decisions in breast cancer management.