Hybrid Medical Data Mining Model for Identifying Tumor Severity in Breast Cancer Diagnosis

Purpose: This study proposes a methodology for detecting tumor severity using data mining of databases relating to breast imaging modalities. In doing so, it proposes creating a software application that can serve as an efficient decision-making support system for medical practitioners, especially those in areas where there is a shortage of modern medical diagnostic devices or specialized practitioners, such as in developing countries. Method: We investigated the data of approximately 3754 screened women by using “BI-RADS” categories as a quality assessment tool to screening, measure, and identify the size and location of lesions, determine the number of lymph nodes, collect biopsy samples, determine final diagnoses, prognoses, and age which were all available from the screening registry. Result: The application of each algorithm on BI-RADS values 4 and 5 for Invasive Ductal Carcinoma lesions was assessed, and the following accuracy was acquired: CART: 84.71%. In order to get the best result, four optimum clusters based on tumor size were applied for constructing simple rules with significant confidence. Conclusion: In this study, we present a hybrid approach - a combination of k-means with GRI and CART decision tree - to better assess breast cancer data sets.