CAD Tool for Breast Cancer Prediction Using Multiple Deep-learning Models

Breast cancer is one of the leading causes of cancer death among women worldwide. It represents a global health concern due to the lack of effective therapeutic regimens that could be applied to all breast cancer patients. Breast cancer treatment decisions rely on clinicopathologic parameters. However, this approach is replete with limitations as it fails to define prognosis uniquely and is not always sufficient to settle unequivocally on the best type of treatment for breast cancer patients. The molecular diagnostic efforts have been focused mainly on Estrogen Receptor (ER)‐positive (Luminal A) breast cancer being the most represented breast cancer subtype (70% of patients) with a standard treatment (endocrine therapy for five years) and a good prognosis. However, at least 20% of patients will suffer a distant recurrence within ten years. Although many molecular tests have been developed to identify the patients at risk of recurrence, a definite, reliable and effective in vitro diagnostic device that stratifies patients at high risk and low risk of relapse, directing therapeutic decisions, is still a significant clinical need. This study aims to fill this gap by investigating and developing a new approach for better stratification of breast cancer patients in the risk categories of recurrence. It is based on the integration of clinical and digital pathology analysis. The combined analysis, indeed, aims to further categorize the patients with an intermediate risk of recurrence either in the low-risk group with no necessity of chemotherapy or in the high-risk group that needs chemotherapy. The paper presents the approach, the implemented Computer-Aided Diagnosis (CAD) tool and finally, the results of evaluating its predictive accuracy. The tool achieved 88% accuracy in histological image classification, 95% in cancer grade prediction and 71% in 10-year recurrence prediction.