Resolution Bounds and Reconstruction Techniques for Time-resolved Transillumination Imaging

Recent technological advances now enable time-gated acquisition of photons at very fast rates. This allows one to separate scattered and unscattered photons by temporal gating, a process termed time-resolved transillumination (TRT) imaging. TRT imaging opens the door to a new type of scanning through turbid (scattering) media such as soft tissue and fog/smoke, and many potential applications in bioimaging and surveillance. This paper investigates the performance tradeoffs between TRT imaging and conventional (un-gated) imaging. On the one hand, time-gated, firstarrival photons are unscattered and therefore provide very high spatial resolution. But, very few photons arrive at the detector without scattering, effectively resulting in a very low SNR. On the other hand, conventional (un-gated) imaging is based on all photons (scattered and unscattered), resulting in lower spatial resolution, but higher SNR (due to the large number of photons). This paper investigates these tradeoffs using a decision-theoretic approach to ascertain bounds on the minimum resolvable occluding object size with and without time-gated photon acquisition. The theoretical predictions are validated through a realistic simulation of tumors in breast tissue. The paper then proposes a novel Maximum Likelihood approach to TRT image reconstruction. Using a novel Expectation-Maximization algorithm, multiple snapshot observations of time-gated photons are used to reconstruct the image. This allows us to obtain the ”best-of-both-worlds” by combining the spatial resolution of the first-arrival photons, with the higher SNR provided by scattered photons.