Machine learning methods based on mammogram images to estimate survival times for breast cancer patients

Abstract

Estimating survival times based on medical images resulting from radiological imaging of some parts of the body affected by tumors and making a predictive system for them is considered one of the very important fields at the present time. This is because radiological imaging is one of the first and most important stages of medical diagnosis. Therefore the process of linking medical images including them within the work steps of the statistical analysis for estimating survival times is a modern and important topic. Its importance is helping doctors and medical specialists to determine the influencing factors and risk percentage associated with survival for each patient based on the medical image of the affected part.
In this paper, the medical images extracted from the mammogram device for breast cancer patients in Iraq. These images were included in the machine learning method for estimating survival of patients. Based on two methods to extract features, The first one is the Fast independent component algorithm (Fast ICA algorithm) and the second one is Nonnegative Matrix Factorization (NMF algorithm). With two machine learning algorithms, The first method is Random survival forests algorithm and the second method is support vector machine algorithm SVM (SVM).
Through the application of the supervised machine learning method on mammogram images of patients with breast cancer, it was found that the best model for estimating survival according to the mean square error (MSE) and concordance index (C-Index) criterion is the model resulting from the use of the Fast ICA algorithm with the random survival forest algorithm compared with the other three models. Accordingly, It is recommended that interested medical agencies and institutions to adopt this model.