Using time-resolved data to predict an object’s form behind a scattering medium, or ballistic photons to image through tissue.
Researchers at MIT have used deep learning on time resolved data from the PF32 camera in order to predict an object’s form behind a scattering surface. The figure below shows (a) the training of the computer neural network (CNN) and (b) the physical setup of the pulsed laser and PF32 camera.
By analysing (a) the time-resolved data from the PF32 of an unknown object form, the authors were able to (b) predict the classification of the toy’s pose with a high degree of accuracy as shown the figure below.
In other work, researchers within the Proteus project have used the PF32 for locating optical endomicroscopy fibres inside organic tissue by using the setup below (a) – (c). The pulsed light that exits the fibre is mostly scattered by the tissue. Some photons, however, pass through without undergoing scattering events. These are known as ballistic photons. In (d) at t=X, we observe these ballistic photons arriving at the camera, at times beyond this the camera receives light that has been scattered within the tissue. Using the arrival time of the ballistic photons with respect to the time at which the laser pulse was emitted, the team can ascertain where the fibre is in the z dimension, with the x and y dimension given by the imaging.