Automated 3D segmentation of hydrogel-treated mice wounds using optical coherence tomography and deep learning

Optical coherence tomography (OCT) is a powerful non-invasive imaging technique, providing high-resolution three-dimensional images of biological tissues. However, accurately segmenting regions-of-interest from large 3D images is challenging and often relies on manual processes that are labor-intensive, subjective, and prone to inter-operator variability. This is particularly true in dynamic processes like wound healing, where tissue structures continuously evolve. This study introduces a U-Net neural network model for automatic segmentation of 3D OCT images from mouse wound models treated with dextran-based hydrogels. The network achieves 85.5% per-pixel validation accuracy in identifying eight structural subtypes, with the same model operating across the entire 14-day healing period. This approach enabled longitudinal in vivo monitoring of hydrogel volume and degradation, providing valuable insights into wound healing dynamics and treatment efficiency without the need for invasive biopsies.