Date of Award

Spring 2020

Document Type



Santa Clara : Santa Clara University, 2020.



First Advisor

Yuling Yan


Cancer is a highly prevalent disease that affects millions of people worldwide. In addition to the physiological effects of the disease, cancer patients are more likely to be diagnosed with Major Depressive Disorder (MDD). Unfortunately, prior research has shown that MDD can also decrease the efficacy of radiotherapy cancer treatments. Currently, there is no way to predict, prevent, or mitigate this comorbidity, preventing physicians from administering supplemental therapies. In this paper, we propose a low-cost and efficient computational tool that can be utilized to quantify a patient’s likelihood of developing depression. To do so, we used PET images and a ResNet34 architecture to train a convolutional neural network to identify depression biomarkers in the brain. These brain PET images were taken from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) dataset and also provided information regarding the patient’s depression at the time of the scan. We were then able to label and classify images in our dataset based off of this data. Although our model only yielded an accuracy of 54.25%, sensitivity of 56.25% and a specificity of 53.64%, a visual evaluation of our results (GradCAM) confirmed that our algorithm was able to detect the correct regions of interest in the brain, where depression biomarkers were found. This leads us to believe that our deep learning model, with improvement, can be used to effectively help classify depression progression rates in radiotherapy patients.