Deep Learning for Improved Clinical Cancer Care
The prostate-specific antigen test (PSA) is a blood test used primarily to screen for prostate cancer. However, single PSA values can be hard to interpret due to noise and fluctuating baselines. This summer, I will be continuing my work at The Hong Lab at UCSF where we are applying deep learning techniques to EHR data in order to better understand PSA growth in potential prostate cancer patients. By taking into account both a patient’s clinical data over time as well as certain demographic features, we hope to build a tool that can accurately predict a patient’s PSA value three months into the future. Further, we hope this research will shed light onto how PSA baselines may vary according to these demographic features so that clinicians can be better informed when making decisions regarding potential cancer cases. To do so, we will be developing a pipeline to best interpret the sparse EHR data that is available and employing complex time series analysis tools, such as Temporal Convolutional Neural Networks, to learn powerful features from our patient specific time series data.
Message to Sponsor
- Major: Applied Mathematics
- Mentor: Julian Hong, UCSF