361 — Generating perfusion maps from structural optical coherence tomography with artificial intelligence

Lee et al (10.1101/271346)

Read on 16 August 2018
#retina  #OCT  #diagnosis  #flow  #turbulence  #ophthalmology  #eye  #blood  #perfusion  #machine-learning  #neural-network 

Much like MRIs, there are many species of OCT. In the past, I’ve written almost exclusively about structural/anatomical OCT but perfusion and flow can also be measured using OCT imagery. (Angiography, measured with OCTA, is different than light-scattering functional OCT, as I discussed here).

In practice, OCTA, which measures blood flow in retinal vasculature, requires different hardware and software from the standard OCT machinery; and so it is less widely used.

These researchers propose a machine learning technique that aims to output results akin to OCTA flow data, using only the anatomical OCT data as an input. The deep learning network only approximates superficial flow map data and ignores deep retinal vasculature, but easily reconstructs those vessels shallower in the retina. This work is perhaps not clinic-ready, but it demonstrates that anatomical OCT scans capture a lot of information about perfusion that machine learning can leverage to create flow maps of the same tissue without rescanning the patient.