Quantum-mimic Optical Coherence Tomography (Qm-OCT) images are cluttered with artefacts - parasitic peaks which emerge as a by-product of the algorithm used in this method. However, the shape and behaviour of an artefact are uniquely related to Group Velocity Dispersion (GVD) of the layer this artefact corresponds to and consequently, the GVD values can be inferred by carefully analysing them. Since for multi-layered objects the number of artefacts is too high to enable layer-specific analysis, we employ a solution based on Machine Learning. We train a neural network with Qm-OCT data as an input and dispersion profiles, i.e. depth distribution of GVD within an A-scan, as an output. By accounting for noise during training, we process experimental data and estimate the GVD values of BK7 and sapphire as well as provide a qualitative GVD value distribution in a grape and cucumber. Compared to other GVD-retrieving methods, our solution does not require user input, automatically provides dispersion values for all the visualised layers and is scalable. We analyse the factors affecting the accuracy of determining GVD: noise in the experimental data as well as general physical limitations of the detection of GVD-induced changes, and suggest possible solutions.
© 2023. The Author(s).