Predicting neoadjuvant treatment response in triple-negative breast cancer using machine learning

Area of Science:

• Oncology
• Computational Pathology
• Biomarker Discovery

Background:

• Neoadjuvant chemotherapy (NAC) is standard for early-stage triple-negative breast cancer (TNBC), aiming for pathological complete response (pCR).
• Current pCR prediction for NAC in TNBC is limited, with only 30%-40% achieving pCR.
• Identifying predictive biomarkers is crucial for stratifying TNBC patients and guiding therapeutic decisions.

Approach:

• Utilized a supervised machine learning (ML) approach to evaluate predictive biomarkers from H&E and IHC stained biopsy tissue.
• Developed mask region-based CNN (MRCNN) models for detecting tumor cells, stromal and intratumoral TILs (sTILs and tTILs), Ki67+, and pH3+ cells from whole slide images (WSIs).
• Identified 'hotspot' regions with high cell densities and trained ML models to predict NAC response using various histological and molecular features.

Key Points:

• Highest prediction accuracy was achieved using hotspot regions defined by intratumoral TILs (tTILs) counts.
• Combining multiple histological features (tTILs, sTILs) and molecular biomarkers (Ki67, pH3) demonstrated top performance at the patient level.
• The complementary use of these biomarkers significantly enhances prediction accuracy for NAC response.

Conclusions:

• Prediction models for NAC response in TNBC should integrate multiple biomarkers rather than relying on single markers.
• Machine learning-based models show significant promise for accurately predicting NAC response in TNBC patients.
• This approach can aid in precise patient stratification, optimizing treatment strategies for TNBC.