Predicting Invasive Disease-Free Survival Time in Breast Cancer Patients Using Graph-based Semi-Supervised Machine Learning Techniques

Breast cancer is currently the most frequently diagnosed cancer and leading cause of cancer death in women worldwide. Researchers have been researching the best treatment for breast cancer. Their focus has been on preventing recurrence after the initial treatment of patients. Choosing the appropriate model for survival time analysis is the main challenge in the survival analysis of these patients. In this applied research, using graph-based semi-supervised learning method a new survival analysis model is proposed for analyzing the survival of breast cancer patients. Also, a dataset of 3833 patients with breast cancer who were followed up for 5 years was used. The clinical and pharmacogenomics information, including the results of Tamoxifen use and effect on the treatment of invasive cancer, was also available and used. To validate the proposed model, by introducing the evaluation parameters and simulation of the models in MATLAB software, the model performance is compared with previous models of survival analysis. The results demonstrate that by applying the proposed model in predicting invasive disease-free survival time and when a combination of the clinical and the pharmacogenomics features were used the estimation accuracy was 14% higher than when only the clinical features were used. Moreover, the estimated accuracy was 15% higher than when only the pharmacogenomics features were used. The proposed survival analysis model has a high capability for identifying survival risk and high accuracy in predicting patients' survival time.