Author(s): Phd C. Charles Bteiche, Prof. Edwin Hancock
Topcon Europe Medical B.V, International Account Manager, Product Specialist
Purpose: This paper describes a novel approach for automatic 3D mesh extraction and quantification in retina swept source oct-angio imaging without human intervention, based on three consecutive novel methods
Methods: FDTR (Filtering/Detecting/Tracking/Rebuilding)
The defined steps are described as a) developing and applying a novel bio-selective filter (Bteiche-Hancock) (b) detect (using deep learning method, features classification) and track neo-vasculature structure using enhanced multiple instances learning temporal tracking algorithm (c) extracting 2D meshwork and build 3D using cubic neighborhood bifurcation and splines fitting technique.
Results: Specificity of retina layers’ filter between IPL and RPE is a must, in order to countermeasure a multiplicative speckle noise while enhancing neo-vascular structure view and detection. Extraction of NVs through temporal tracking can be achieved using an adapted multi-instance learning tracker through retina multi-frames.
3D structure reconstruction can be operated by extrapolating the skeletonizing/neighborhood detection of bifurcation; then apply splines fitting to smooth the results.
Conclusions: Detection and extraction of non-uniform structures within the retina layers will open widely the door to 3D structural clinical analysis and achieve faster and more accurate diagnosis for retinal diseases through the different layers of the retina. This step will empower the clinician to understand more the progression a disease while comparing its progression in 3D and will open widely the door for more comprehensive data collection for AI to conquer the early detection and progress of diseases.
Financial Disclosure: Phd research subjected is totally funded by the author, neither Topcon nor other companies have any financial or non-financial contribution in the research underwent.