Dissertation Talk: Machine Learning and Robotics in MRI: From Image Reconstruction to In-Scanner Intervention

Event details Date: Thursday, August 6, 2026 Time: 2:00 PM to 2:00 PM Location: 490 Cory (Swarm Lab, Immersion Room) Type: Performing Arts Audience: Faculty,Students About this event Magnetic Resonance Imaging (MRI) is one of the most capable measurement instruments in medicine, but it is relatively slow and imposes substantial constraints on devices operating inside the scanner. Faster acquisitions can shorten scan time and reduce motion artifacts, including those caused by breathing, but can be more sensitive to physical imperfections. Reconstruction can mitigate these imperfections after data acquisition, but an end-to-end approach to MRI systems requires more than reconstruction: feedback can act upstream by shaping patient breathing, while MRI-compatible localization and actuation can extend the scanner from a measurement system toward a platform for intervention. This dissertation follows that progression from robust image formation, through physiological guidance, to in-scanner device localization and actuation. In this talk, I present three contributions that address these challenges. The first comes from work on learning-based reconstruction: a physics-informed framework that corrects off-resonance blurring in fast non-Cartesian MRI. Trained entirely on synthetic noise-like data, it generalizes across anatomies and contrasts without retraining. The second is an adaptive soft robotic vest that combines real-time sensing with timed pressure feedback to guide breathing toward more regular and consistent patterns during MRI and other motion-sensitive procedures. The third integrates robotic positioning of a Transcranial Magnetic Stimulation (TMS) coil inside the scanner. Custom MRI-compatible hardware and software enable fast localization and closed-loop TMS coil placement. Anatomy, functional targets, and coil pose share a common MRI-defined frame, supporting accurate and verifiable positioning during concurrent TMS, EEG, and fMRI. Across the work, the physics and constraints of MRI shape hardware, acquisition, and computation... Official event details: https://events.berkeley.edu/eecs/event/325217-dissertation-talk-machine-learning-and-robotics-in