Deep-learning-based approach for strain estimation in phase-sensitive optical coherence elastography.

Abstract

In this Letter, a deep-learning-based approach is proposed for estimating the strain field distributions in phase-sensitive optical coherence elastography. The method first uses the simulated wrapped phase maps and corresponding phase-gradient maps to train the strain estimation convolution neural network (CNN) and then employs the trained CNN to calculate the strain fields from measured phase-difference maps. Two specimens with different deformations, one with homogeneous and the other with heterogeneous, were measured for validation. The strain field distributions of the specimens estimated by different approaches were compared. The results indicate that the proposed deep-learning-based approach features much better performance than the popular vector method, enhancing the SNR of the strain results by 21.6 dB.