A Deep Learning Pipeline for the Segmentation of In Vitro Wound Healing Microscopy Images following Laser Therapy

Disruption of the epithelial tissue's integrity leads to wounds. Damaged tissue raises the possibility of infection, which has a negative impact on many aspects of life. Recent interest has focused on several novel approaches to improve quality of life, including photobiomodulation (PBM) research that emphasizes wound modeling. The contribution of the PBM method to the wound healing process is examined by in vitro studies. The size of the area recovered from the microscopic examination images is the primary success criterion in wound healing research, which investigates the effectiveness of various parameters such as the laser wavelength, power, and exposure duration. Therefore, segmentation is a crucial step in analyzing obtained images and has a significant role in conducting accurate analysis. In this study, a U-net structure-based deep learning (DL) approach was presented for accurately segmenting microscopic wound healing images from PBM studies. The success of the developed DL model was evaluated with various performance metrics and compared with ground truth labels, which were manually determined by a blind expert. Most of the performance metrics utilized had success rates of over 90 %. The average dice similarity coefficient (DSC) between ground truth labels and the DL model's prediction was obtained as 0.953 and 0.939 for the validation and test image sets, respectively.