Justin Kim Rose Hills
GANs for Predicting Growth in Subsolid Pulmonary Nodules From CT
Lung cancer is the leading cause of cancer death worldwide, and catching it early saves lives. My research focuses on subsolid nodules, small growths on CT scans that can be early lung cancer precursors. They are difficult types of nodules for radiologists to accurately assess, as many are infections that resolve, but others carry malignancy risk, requiring monitoring over months or years of scans–a burden we’d like to avoid for patients whose nodules resolve on their own. This project will explore a deep learning framework called a Generative Adversarial Network (GAN) for this problem. Image-to-image GANs learn fine-grained features in images to generate predicted future CT imaging of each nodule so potential growth and risk can be assessed from a single scan.
My proposed model, trained on a large-scale imaging database, could help clinicians identify potentially dangerous nodules earlier and ease the imaging burden on patients and hospitals. That matters especially for underprivileged communities who face more difficulty accessing advanced imaging to monitor these nodules, and for whom faster, less labor-intensive care could make a difference.
Message To Sponsor
Thank you so much for your generosity and support in making this fellowship possible. This opportunity means a great deal to me, both as a chance to grow as a researcher and to contribute to a cause I care about. I'm grateful to be able to spend this summer developing machine learning approaches that I hope can eventually grow into tools that improve patient outcomes at scale.