Survival Outcome Prediction for Stereotactic Body Radiation Therapy of Lung Cancer from Post-RT Ct Images with RNN/CNN Deep Learning.

Radiation therapy has shown to be an effective method for treatment for non-small cell lung cancer. Post-treatment follow-up CT scans are a vital non-invasive method in tracking the tumor response to treatment and detecting early signs of recurrence or radiation induced lung injury. Additionally, following the radiographic characteristics of a treated tumor over time can provide insight and prediction regarding outcome for a patient, more specifically post-treatment survival time. In this paper, we propose a patient-specific model based on Deep Convolutional and Recurrent Neural Networks to track a treated tumor over time making a two-year post-treatment survival time prediction. We use a cohort of data from Wake Forest University that consists of pre and post radiation-treated lung cancer patients with a wide range of follow-up scans. We train our model with these data and demonstrate that our patient specific model can predict two-year survival outcome with an ROC-AUC score of 0.718 on our hold-out test set.