Artificial Intelligence in Breast Cancer MRI: Applications for TNM staging

Breast cancer (BC) is the most common non-cutaneous malignancy in women: a woman living up to the age of 90 has a one in eight chance of developing breast cancer. Currently, we know that BC is a very heterogeneous pathology, with a great variety of presentation and clinical, anamnestic and histological scenarios that determine important phenotypic variability. It is therefore essential to identify the details of the tumor and make a real "identikit", in order to guide the therapeutic path and accurately establish the prognosis Through invasive procedures we obtained tumor histological prognostic factors; while non-invasive procedures, as medical images, allow us to investigate tumor characteristics but also to determine the correct TNM (Tumor, Nodes, Metastases) staging. The BC loco-regional staging is performed with pre- and post-contrast Magnetic Resonance Imaging (MRI) sequences. In recent years, pilot studies have attempted to correlate the tumor characteristics extracted from MRI with the molecular subtypes of BC through artificial intelligence systems, with the aim of predicting their biological behavior. This thesis concerns the works we published in about 3 years on the role of artificial intelligence in the evaluation of the breast cancer TNM through MRI. Nevertheless, we dealt with three major topics: the BC primary tumor, the axillary lymph node involvement and the distant metastasis. We conducted reviews about the role of AI in lymph node involvement and bone metastatic BC localization predictions, and proposed new Radiomics and DL approaches in order to evaluate the performance in the prediction of the three main investigated topics: BC local aggressiveness, LN involvement and distant metastatic spread.