Fast Intelligent Autofocusing for Retinal Optical Coherence Tomography
Pablo Ortiz
pablo.ortiz@duke.edu
| Paper PDF |
|
Our project aims to improve the focusing speed of retinal optical coherence tomography systems as well as automate the procedure. Current auto-focusing techniques are slow, so they cannot be applied in real time and therefore cannot be executed during image acquisition--it has to be done beforehand. We trained a neural network to evaluate, from a single b-scan, the level of defocus present in the system. This way we are able to control the focus level on real time by correcting for it in less than 0.3s. without the need to interrupt the image acquisition. See this video to observe the system working in real time. The video shows live-oct volumes. Defocus is introduced 4 times during the video, but the system corrects for all of them very quickly. |
|
|
| Paper: |
Code and Data:
|