Accurate risk assessment of disease recurrence in women with breast and ovarian cancer is crucial for determining which individuals will benefit from systemic therapy. Currently, treatment decisions for these women are often based on personalized molecular tests, which have been shown to improve the prediction of clinical outcomes and have been validated as prognostic or predictive biomarkers. However, there is still room for improvement in risk assessment for these individuals. To address this issue, I plan to use deep learning to optimize treatment decisions at two levels. Firstly, my aim is to use medical image analysis and computer vision algorithms to accurately quantify morphological parameters, such as the tumor-stroma ratio by segmenting relevant tissue groups on hematoxylin and eosin (H&E) stained slides of breast resections. This parameter is thought to be predictive of clinical outcomes, but its implementation in pathology reports is rare due to difficulties in extraction and inter-rater variability. Consequently, I aim to improve risk assessment in early stage breast cancer patients in order to identify individuals who may not need systemic therapy (adjuvant chemotherapy) or who would benefit the most from it.