Model to Predict the Risk of Breast Cancer

About

Introduction: Ductal carcinoma in situ (DCIS) is the accumulation of abnormal cells inside a milk duct in the breast. It is non-invasive and is considered as stage 0 breast cancer, meaning that the cancer cells have not spread through the walls into the nearby breast tissues. According to the American Cancer Society and National Institutes of Health, about 1 in 5 new breast cancers will be DCIS. In 2020, over 1 million U.S. women were affected by DCIS. After being treated with lumpectomy, DCIS patients still need further evaluation and even need to consider additional therapies. One of the goals of DCIS treatment is to prevent local (especially invasive) recurrence. However, current methods of predicting recurrence have had inconsistent results, mainly due to human eye limitations and a lack of consistency among how doctors assess risk. This can lead to uncertainty in therapeutic decision-making, as well as over- or under-treatment. Thus, there remains an unmet need for accurately predicting the recurrence risk for DCIS patients treated with lumpectomy. Technology: Georgia State researchers have developed a novel machine learning-based image analysis pipeline to predict the 10-year risk of ipsilateral recurrence using digitized whole slide images (WSI) and clinicopathologic long-term outcome data collected from lumpectomy-treated DCIS patients. The researchers identified prognostically relevant features from the surgical tissue slides and designed the classification process to predict the risk. A combination of eight features was used to generate a recurrence classifier, and results have shown that this model has significantly predicted recurrence risk in two independent patient groups. It was also shown that this model outperformed traditional histopathological variables in some of the traditional performance metrics, such as accuracy, specificity, positive predictive value (PPV), negative predictive value (NPV), and odds ratio (OR) in this study.