Deep learning-based boundary detection and compensation technique for the accurate flow measurement near the vessel wall and fluid–structure interaction

An accurate analysis of fluid–structure interaction (FSI) at compliant arteries via ultrasound (US) imaging and numerical modeling is a limitation of several studies. In this study, we propose a deep learning-based boundary detection and compensation (DL-BDC) technique that can segment vessel boundaries by harnessing the convolutional neural network and wall motion compensation in near-wall flow dynamics. The segmentation performance of the technique is evaluated through numerical simulations with synthetic US images and in vitro experiments. The neural network algorithm performs well with a high structural similarity of over 0.9 and 0.82 for lumens and vessel walls, respectively. Then, the performance of the wall motion compensation is examined with or without applying DL-BDC to in vitro compliant phantoms. When DL-BDC is applied to flow influenced by wall motion, a bias error is less than 0.0002 cm/s compared with those of the theoretical values. The technique is utilized to analyze FSI with varying elasticities of the phantom. Results show that the flow dynamics and wall shear stress values are consistent with the expected values of the compliant phantoms, and their wall motion behavior is observed with pulse wave propagation. This technique is applied to US images of the human carotid artery to verify the clinical applicability of DL-BDC in analyzing FSI. As a result, DL-BDC can provide fast, accurate, and robust segmentations of vessel walls and high accuracy in near-wall flow measurement by compensating the wall motion. This proposed approach can be beneficial in the FSI analysis of various biomedical applications.