Brian Kunzang Rose Hills
Predicting Growth of Subsolid Lung Nodules Using Deep Learning
Subsolid nodules are a class of lung growths associated with an aggressive form of lung cancer called adenocarcinoma. Such subsolid nodules tend to grow slowly, and some can gradually progress to invasive lung cancer, requiring repeated CT scans over many years to determine potential malignancy. As a result, patients often receive large doses of radiation through CT scans, incur high healthcare costs, and experience heightened anxiety.
My project will focus on applying modern deep learning methods to this problem to create a model that can classify whether a nodule will grow or not based only on an initial scan, as opposed to tracking it over many follow-up appointments. Such approaches have found success in areas of medical imaging, but there has been comparatively little work on deep learning in relation to subsolid nodules. To this end, I plan to train a convolutional neural network on a large dataset of CT images of subsolid nodules that I have helped process at the Sohn Lab. Such a model that accurately predicts nodule growth would help clinicians detect cancer at an even earlier stage than they would be able to otherwise.
Message To Sponsor
Thank you for supporting my project! I am deeply grateful for the opportunity to focus fully on my research this summer. As a first-time undergraduate researcher, I am excited to explore deep learning and its applications in radiology as I take an initial step toward a research-focused career.