Applying Machine Learning on Breast Mass Digitized Images to Predict Breast Cancer

Breast cancer develops in the glandular tissue of the breast affecting the cells that line the lobules. Early and accurate detection and receiving cancer treatment at earlier stages can prevent many deaths among millions of women across the world. The histological evaluation of a tumor generates a tremendous amount of data related to malignancy, but findings may vary depending on subjective analyses. Therefore, a computer-based diagnostic system is required for accurate diagnosis and early detection. This study analyzes breast cancer diagnostic dataset from the Wisconsin Breast Cancer Database using the IBM Watson Machine Learning Platform. The dataset contains 569 digitized images of individual breast masses (357 benign, 212 malign) with cell nuclei information. The dataset captured a range of properties, including radius, texture, and perimeter, that were employed as variables influencing the outcome. IBM Watson machine has the capability of developing models by using multiple algorithms. In this study, two algorithms with various enhancements were used including Snap Logistic Regression and Snap SVM Classifier with enhancements. Both algorithms reached an accuracy of about 97%. Some essential image features including the product of various factors like area-worst, concave points, symmetry worst, smoothness mean, and perimeter worst were found playing a significant role in predicting the cancer cells malignancy.