Correlation Study Between Neoadjuvant Chemotherapy Response and Long-Term Prognosis in Breast Cancer Based on Deep...

Area of Science:

• Oncology
• Computational Biology
• Medical Informatics

Background:

• Pathological response to neoadjuvant chemotherapy (NAC) is crucial for breast cancer outcomes.
• Current binary assessment (pathological complete response) misses prognostic details across subtypes.
• Need for advanced models to predict recurrence and metastasis after NAC.

Purpose of the Study:

• Develop an interpretable deep learning model for breast cancer recurrence prediction post-NAC.
• Integrate diverse clinical and pathological variables for enhanced prognostic accuracy.
• Improve risk stratification beyond conventional pathological complete response evaluation.

Main Methods:

• Retrospective analysis of 832 breast cancer patients treated with NAC (2013-2022).
• Utilized Multi-Layer Perceptron (MLP) model with variables: tumor size change, nodal status, Ki-67, Miller-Payne grade, and molecular subtype.
• Benchmarked MLP against SVM, Random Forest, and XGBoost using cross-validation and performance metrics (AUC, accuracy, precision, recall, F1-score).

Main Results:

• MLP model achieved high AUCs: 0.86 (HER2+), 0.82 (TNBC), 0.76 (HR+/HER2-).
• SHAP analysis revealed post-NAC tumor size, Ki-67, and Miller-Payne grade as key predictors.
• Patients achieving pathological complete response still faced a 12% recurrence risk, underscoring limitations of binary assessment.

Conclusions:

• The deep learning system offers precise, interpretable risk assessment for NAC patients.
• Facilitates personalized treatment strategies and tailored post-treatment monitoring plans.
• Highlights the need for continuous risk evaluation beyond initial response assessment.