Using Machine Learning to early detection and classification of breast cancer masses based on medical image processing

Abstract

Abstract In this paper, a link was made between one of the machine learning methods, which is Artificial Neural Networks (ANN), and the analysis of Medical Images (mammography images) for the classification and early detection of breast cancer (the most prevalent among women in the world). The idea of ​​the research depends on the Feature Extraction of some measurements (statistical and geometrics) of the shape of benign and malignant masses using the digital image processing program (ImageJ) and then using the method (Feed Forward Networks) to classify the process between them for the data of a sample of patients with (150) digital mammography images (75 benign and 75 malignant images) in breast cancer. The research reached a classification accuracy of (86.67%) Model Sensitivity equal to (91.67%) to distinguish between the two types masses, which is a high percentage based on a set of statistical criteria for evaluating the model (confusion matrix, ROC curve, Kappa statistics) The variables (Perimeter, Minor, InDen, and Feret) were ranked first in importance and distinction between the two masses, and the program (R4.2.3) was used to analyze the data.