Breast Cancer Survival Prediction Modeling Based on Genomic Data: An Improved Prognosis-Driven Deep Learning Approach
Amena Mahmoud, Musaed Alhussein, Khursheed Aurangzeb, Eiko Takaoka
Abstract
Breast cancer has a wide range of possible outcomes due to its complexity and heterogeneity. The process of manually detecting breast cancer is laborious, intricate, and inaccurate. It is essential for individualized treatment planning to have a reliable prognosis of patient survival. Increased focus in recent years has been placed on genomics-based techniques be-because of their potential to better predict outcomes. In this study, we propose a novel framework for breast cancer survival prediction using optimized deep learning models. We begin by preprocessing and integrating multi-omic data, including gene expression profiles, somatic mutations, and clinical features, obtained from a large cohort of breast cancer patients. In our proposed research, deep learning models were trained to detect the survival case of breast cancer and were optimized using Stochastic Gradient Descent Optimizer which was used for the initial population generation and modification for the selected dataset and divided into 80% for the training set and 20% for the testing set. Long Short-Term Memory, Variational Autoencoders, and Graph Convolutional Networks architectures optimized by Stochastic Gradient Descent Optimizer are used for training and validation of the breast cancer dataset and get the best accuracy of 98.7% for the optimized Long Short-Term Memory model. Our results demonstrate that the proposed genomics-based predictive modeling approach achieves high performance in breast cancer survival prediction compared to conventional methods.