Dynamic Numerical Biomarker Classification Using Recurrent Neural Networks to Predict Tumor Downstaging and Neoadjuvant Chemotherapy Efficacy

• Sunday Sunday

  Ladoke Akintola University of Technology

  Author

Neoadjuvant chemotherapy (NAC) has become a cornerstone in the management of locally advanced and high-risk early-stage cancers, yet predicting individual patient response remains a significant clinical challenge, as pathological complete response (pCR) is achieved in fewer than 30% of patients. Current predictive approaches predominantly rely on static, single-time-point imaging or histopathological features, failing to capture the temporal dynamics of tumor evolution during treatment. This study addresses this gap by developing a dynamic numerical biomarker classification framework employing Recurrent Neural Networks (RNNs) to predict tumor downstaging and NAC efficacy. A retrospective cohort of 463 breast cancer patients was analyzed, with longitudinal numerical biomarker data—including C-reactive protein (CRP), tumor mutation burden (TMB), lactate dehydrogenase (LDH), and serial imaging-derived quantitative features—extracted at multiple time points throughout the NAC cycle. The proposed RNN-based sequential model, augmented with residual and inception layers, achieved a validation accuracy of 89.4% with an ROC-AUC of 0.8955, significantly outperforming static machine learning baselines including Random Forest (85.0%) and XGBoost (82.8%). Feature attribution analysis identified dynamic trajectory patterns of CRP and TMB as the most significant predictors of pCR, with contribution scores exceeding 15%. The framework demonstrates that temporal biomarker evolution provides critical predictive information absent in static assessments, offering clinicians a non-invasive, cost-effective tool for early treatment adaptation. This study establishes the feasibility of dynamic numerical biomarker classification using RNNs and provides a replicable framework for precision oncology decision support, with implications for personalized chemotherapy optimization and improved patient outcomes.

• PDF

Copyright (c) 2026 Sunday Sunday (Author)

This work is licensed under a Creative Commons Attribution 4.0 International License.