Neoadjuvant Chemotherapy (NACT) is increasingly used for pre-operative treatment of breast cancer patients. Successful application of NACT, resulting in a substantial or complete reduction in tumor volume, enables breast-conserving surgery in a higher number of cases. In addition, NACT allows assessment of tumor sensitivity to chemotherapy. The most reliable measure of NACT effectiveness is quantification of post-operative residual disease via histology, which has been shown to be a strong indicator of long-term prognosis. Although many patients have indeed substantial benefit from neoadjuvant treatment, there is a large group of patients not responding while still experiencing the toxic side effects. To date, it is impossible to predict upfront whether a patient will respond to NACT.
There is existing evidence of some correlation between histopathological features (e.g. intra-tumoral lymphocyte density, macrophage influx) and NACT response. It is very likely that additional yet unknown features of the tumor cell compartment combined with the tumor micro environment (TME) as a whole exist, which are responsible for treatment response. We therefore conclude that a rich amount of information can be found in histopathology tissue samples, which is currently not utilized to predict treatment response. We make the unique assumption that mining big preoperative histopathological data by artificial intelligence gleans unknown features and can discover new quantifiable visual substrates related to NACT response. We therefore hypothesize that AI-based assessment of H&E-stained digitalized histopathology images of core biopsy of breast cancer will answer the question “whether a patient will respond to NACT”.
The aim of PROACTING (PRedicting neOAdjuvant Chemotherapy Treatment response with deep learnING) is to build a predictive model that will be able to process a digitized histopathology image of breast cancer biopsy and predict the treatment response in terms of pCR. This project will be developed in a period of 2 years as a collaboration between the Netherlands Cancer Institute (NKI) and the Computational Pathology group of the Radboud University Medical Center (Radboudumc).