Triple-Negative, Digital Biomarkers, Survival Analysis, Neoadjuvant Chemotherapy, Histology Images Analysis

Abstract

Timely detection, precise diagnosis, and effective risk stratification play a pivotal role in optimising treatment decisions and enhancing outcomes for individuals battling cancer. The advent of Digital Pathology (DP) introduces a revolutionary potential to elevate cancer detection methods and refine treatment management by improving diagnostics and prognostics. This thesis endeavours to harness the power of deep-learning techniques for analysing histology images in triple-negative breast cancer (TNBC), with the ultimate goal of extracting digital biomarkers to enrich the study of patients' outcomes.

The analysis of whole slide images commences with a robust tissue classification, followed by intricate computational examinations to extract spatial features of the tumour microenvironment. This inquiry unveils correlations and relationships between the studied features, patients' treatment responses, and survival outcomes. The refined tissue classification model emphasises the significance of tumour-associated stroma and stromal tumour-infiltrating lymphocytes in predicting patients' responses to neoadjuvant chemotherapy. Spatial quantitative measures derived from the computational analysis serve as invaluable digital biomarkers, providing crucial insights into risk outcomes for individuals with TNBC.

Moreover, delving into NanoString data and exploring digital basal and non-basal subtyping of TNBC extends the scope of this thesis and augments the comprehension of the disease. This broadening of perspective opens avenues for potential connections among histopathological characteristics, molecular profiles, and disease subtypes, thereby enhancing the prospects for personalised treatment strategies to advance.