Proposing a Clinical Decision Support System for Breast Cancer Diagnosis

Breast cancer is one of the leading causes of death in women. Therefore, the accuracy and speed of diagnosis are crucial in the treatment procedure. In this regard, the BI-RADS classification system has been used for the standardization of mammography reports. However, there is much disagreement among physicians about the BI-RADS values. The aim of this paper is to diagnose BI-RADS by natural language processing of mammography reports and clinical information from the electronic health records and combining them to identify molecular subtypes and help patient follow-up.

In this study, 1200 mammography reports and electronic health records obtained from Namazi Educational and Medical Center for years between 2015-2017. After text processing, the vector with 160 features was obtained, then 18 features were extracted by referring to the electronic health records. Finally, 178 features were used by SVM and naïve Bayesian to predict BI-RADS and molecular subtypes, respectively.

The values of Accuracy, Positive Prediction Value, Negative Prediction Value, Sensitivity, and Specificity were calculated to evaluate the results. Accuracy was 85.42% for BI-RADS and 72.31% for molecular subtypes.

The proposed decision support system was an appropriate model to help the physician to diagnose breast cancer and categorize patients. It was also found that the combined information, including electronic medical records of patients and designated molecular subtypes along with mammography reports, can be useful in diagnosing the disease and defining the treatment follow-up