Deep learning for residual disease stratification in early breast cancer

Thesis

Breast cancer remains the leading cause of cancer-related deaths among women globally. Despite advances in treatment, the decision-making process for administering neoadjuvant systemic chemotherapy (NAC) still poses significant challenges. This research aims to improve the characterisation of residual disease (RD) in surgical specimens of breast cancer patients following NAC. Utilising a deep learning approach to analyse histopathological information from Whole Slide Images (WSIs), this study seeks to predict overall survival (OS) and disease-free survival (DFS) more accurately. While RD is a known prognostic factor, its presence does not unequivocally predict patient outcomes, as evidenced by variations in patient response to NAC. Our study will focus on the PRIMUNEO dataset, encompassing 500 patients from various cancer centres, and will later extend to the CGFL dataset for external validation. The goal is to enhance the stratification of breast cancer treatment, particularly in the post-neoadjuvant setting, considering the direct assessment of chemosensitivity and the need for potential treatment adjustments based on RD.

Authors: Morel, L

• DOI: 10.5287/ora-qyaa1pko2
• Type of award: DPhil
• Level of award: Doctoral
• Awarding institution: University of Oxford
• Language: English
• Keywords: neoadjuvant; digital pathology; computational pathology; oncology; chemosensitivity response; breast cancer; deep learning; histology