Abstract: Exponentially increasing clinical data are ideal for applying artificial neural networks (ANNs) to predict disease
progression and outcomes. We trained an ANN to predict survival outcome using The Cancer Genomes Atlas (TCGA)lung cancer clinical data . Using a weighted cost function, ReLU activation function, and adaptive gradient descent1) speeded up the training, 2) reduced the training epochs, and 3) predicted the survival outcome with 81% precision and 54% recall on the test set.
Learning Objective 1: Our objective is to learn how network behave and predict under different configurations when tasked to predict survival status of patients with only incomplete clinical data.
Kedi Zheng (Presenter)
Wake Forest University
Sam Cho, Wake Forest University
Umit Topaloglu, Wake Forest School of Medicine