Publications
Research at the intersection of machine learning, mathematical optimization, and medical imaging — designing and evaluating multiplexed SPECT systems through physics simulation, Bayesian optimization, and artifact quantification.
Google ScholarConnects detector geometry directly to image quality by deriving pixel-wise tomographic resolution from projection probability density functions (PPDFs) — an analytical ray-tracing model that produces a per-pixel resolution map, so system performance can be evaluated quantitatively before any hardware is built.
Casts SPECT detector design as a search over a parameterized geometry space: analytical ray-tracing yields per-pixel resolution, angular-sampling-completeness, and sensitivity maps for each candidate design, and a joint index (sensitivity × completeness / resolution²) ranks them — replacing trial-and-error engineering with metric-driven optimization.
Multiplexed pinholes share sampling across pixels, so activity at one location can geometrically "ghost" into symmetric locations in the reconstruction. We define a multiplexing similarity index over the system's projection sampling patterns — the fraction of projection beams shared between two pixels — turning ghost-artifact risk into a quantitative, per-pixel geometry metric computable before any data is acquired.