Deep Learning-Based Inclusion Boundary Identification Using Wave Propagation In Optical Coherence Elastography

TY - JOUR

T1 - Deep Learning-Based Inclusion Boundary Identification Using Wave Propagation In Optical Coherence Elastography

AU - Zhang, Guangyu

AU - Liao, Jinpeng

AU - Feng, Zhengshuyi

AU - Layton, Alison

AU - Li, Chunhui

AU - Huang, Zhihong

N1 - Publisher Copyright: © 2026 The Author(s). Journal of Biophotonics published by Wiley-VCH GmbH.

PY - 2026/8

Y1 - 2026/8

N2 - Elastography is widely used to estimate tissue mechanical properties based on wave propagation characteristics. However, a major limitation is that elastography maps often fail to accurately delineate internal boundaries within heterogeneous tissues. To address this issue, we propose a novel deep learning–based framework, termed Boundary Detection Network (BDNet). Depth-resolved phase images capturing wave propagation are acquired using Optical Coherence Elastography (OCE). The proposed method is first evaluated on agar phantoms. Experimental results show that all evaluated models achieve comparable performance on single-boundary samples, with a mean absolute error (MAE) of 2 pixels. For dual-boundary phantoms, BDNet achieves an overall MAE of 4.33 pixels. Furthermore, on a human acne dataset, the proposed method provides the best performance, with an MAE of 19.11 pixels. The proposed BDNet demonstrates high accuracy and computational efficiency, providing a promising tool for assisting abnormal tissue morphology identification in clinical elastography applications.

AB - Elastography is widely used to estimate tissue mechanical properties based on wave propagation characteristics. However, a major limitation is that elastography maps often fail to accurately delineate internal boundaries within heterogeneous tissues. To address this issue, we propose a novel deep learning–based framework, termed Boundary Detection Network (BDNet). Depth-resolved phase images capturing wave propagation are acquired using Optical Coherence Elastography (OCE). The proposed method is first evaluated on agar phantoms. Experimental results show that all evaluated models achieve comparable performance on single-boundary samples, with a mean absolute error (MAE) of 2 pixels. For dual-boundary phantoms, BDNet achieves an overall MAE of 4.33 pixels. Furthermore, on a human acne dataset, the proposed method provides the best performance, with an MAE of 19.11 pixels. The proposed BDNet demonstrates high accuracy and computational efficiency, providing a promising tool for assisting abnormal tissue morphology identification in clinical elastography applications.

UR - https://www.scopus.com/pages/publications/105048061627

U2 - 10.1002/jbio.70345

DO - 10.1002/jbio.70345

M3 - Article

SN - 1864-063X

VL - 19

JO - Journal of Biophotonics

JF - Journal of Biophotonics

IS - 8

M1 - e70345

ER -