Bioimaging, at its essence, measures life-systems onto two-dimensional (2D) sensors. However natural signals –with three-dimensions for space, one for time, one for electromagnetic spectrum, etc.– are multi-dimensional (nD). How do you design and build optical systems with  2D sensors to acquire these nD signals? Traditional answer is to simply measure only the dimensions we think are informative. 

 

In Wadduwage lab, we take a different approach. Using data and modern machine-learning tools, we “learn” an informative 2D representation of the nD signal constrained to the physics of light. To do this, we need: (0) data; (1) machine learning algorithms that understands the physics of light; (2) processes to fabricate the unconventional optics designed by the algorithms; (3) physical instruments with the fabricated optics; and finally (4) computational algorithms to extract information from the measurements.

 

So, along with our collaborators we:

(0) Collect large datasets of living-systems,

(1) Build a new breed of wave-physics-informed neural network models and optimizers,

(2) Design and fabricate customized (or “learned”) optics,

(3) Build new optical systems with “learned” optics, and

(4) Utilize state-of-the-art neural networks from computer vision to extract data from the optical systems.