Project summaries are below. Many projects include links to associated publications and source code repositories.
Real-Time Lung Motion Tracking using Deep Learning
Accurate measurement of tumor position is critical for radiation therapy treatment. A deep learning strategy is used to localize tumor position in 3D using a single 2D projection x-ray image, which can be acquired in real-time during treatment. The patient-specific deep learning model can be trained on a GPU system using PyTorch in a few hours, before treatment occurs.
Sparse Matched Filter for Methane Detection from Imaging Spectroscopy
Methane, a potent greenhouse gas, can be detected at leaks and emission sites using imaging spectroscopy instruments, including NASA/JPL's AVIRIS-NG. This project presents a more accurate and faster detection algorithm, capable of processing an entire months-long campaign dataset in 8 hours with GPU acceleration. Simulated methane plumes were used to validate the algorithm's accuracy.
Radiation Oncology Dose Tracking for Head and Neck Cancers
Soft tissue changes over weeks of treatment directly change the delivered radiaton dose patterns. This project seeks to understand the variability in delivered dose to both tumor and nearby healthy organs. Retrospective analysis of patients who recieved daily fan-beam CT scans for image guidance provides a unique dataset to study dose variability. Analysis and image processing is automated within the RayStation treatment planning system for consistency and time management.
Low-Rank Motion Estimation for 4D Lung Computed Tomography Images
4D (3D+time) CT scans of lung cancer patients present an inverse relationship through the inhale and exhale of the respiratory cycle. This project introduces a multi-scale image registration algorithm that constrains the CT image deformations to be related as a group to represent the respiratory cycle, instead of individual pair-wise image registrations. The algorithm is implemented for GPU acceleration using PyCA.
MRI to Histology Registration Workflow
End-to-end workflow to registration of microscopic histology images to in vivo MR imaging validates quantitative MR imaging biomarkers. These results can be used to improve treatment assesment of non-invasive Magnetic Resonance guided Focused Ultrasound ablation therapies.