A. García-Galindo, M. López-De-Castro, R. Armañanzas
Neoadjuvant therapy (NAT) is the preferred treatment prior to surgery for breast cancer, aiming to shrink tumors. However, patient conditions and clinical factors influence the tumor's response. To enhance personalized care plans, it is crucial to develop predictive tools for patient response to NAT. Machine learning have demonstrated potential in breast cancer prognosis by integrating various modalities. In this study, we propose a cascading approach that predicts NAT response in two stages. First, a model is trained using pre-treatment dynamic contrast-enhanced MRI data. Then, a second model incorporates molecular biomarkers. To identify patients with uncertain predictions, we integrate the Conformal Prediction framework into the non-invasive model, referring them to the second model that includes invasive data. This approach reduces unnecessary biopsies. We explore different algorithms using a clinical dataset, demonstrating the potential of our tool in real-life scenarios.
Palabras clave: Breast cancer, Neoadjuvant therapy, Pathological response, DCE-MRI radiomics, Molecular Biomarkers, Conformal Prediction
Programado
Bioestadística
8 de noviembre de 2023 17:20
HC4: Sala Sacristía