OCT-specific signal features for semi-automatic semantic scans annotation and segmentation

Computer vision approaches have grown exponentially in recent years. Training AI models often requires annotated data. To increase effectiveness of this procedure one can use semi-automatic semantic annotation tools where some simplified approaches (based either on some pretrained models or visible features parameters) are implemented and manually tuned to isolate specific objects. OCT-signals contain information-bearing specific speckle structure and signal attenuation patterns. The parameters of these patterns corresponds to tangible tissue properties (such as scatterers spatial distributions), therefore can be used to construct semi-automatic semantic annotation tools. Using OCT-signal simulation approaches we evaluate the parameters of speckle patterns and attenuation coefficients and propose novel semantic annotation tools for OCT scans. We demonstrate the performance of semi-automatic 3D segmentation and annotation. This tool can be used as a supportive tool for AI applications as well as independent tool for semi-automatic scans segmentations and further characterization.