Deep learning automatically assesses 2-µm laser-induced skin damage OCT images.

Area of Science:

• Biomedical Optics
• Medical Imaging
• Computational Biology

Background:

• Assessing laser-induced tissue damage is crucial for understanding laser-tissue interactions.
• Current methods for evaluating skin damage can be invasive or lack quantitative precision.
• Optical coherence tomography (OCT) offers high-resolution cross-sectional imaging capabilities.

Purpose of the Study:

• To develop and validate a noninvasive, automated method for analyzing 2-µm laser-induced skin damage.
• To quantitatively assess the biological effects of varying laser irradiation doses on mouse skin.
• To compare the performance of different deep learning models for damage segmentation and volume quantification.

Main Methods:

• A mouse skin damage model was created using different doses of 2-µm laser irradiation.
• In vivo imaging of damaged skin was performed using optical coherence tomography (OCT).
• Deep learning models (U-Net, DeepLabV3+, PSP-Net, HR-Net) were trained for image segmentation and damage volume quantification.

Main Results:

• HR-Net exhibited the best performance among the evaluated deep learning models.
• HR-Net achieved the highest agreement between segmented and actual damage volumes with minimal error.
• A dose-dependent increase in skin damage volume was observed, with specific volumes correlated to irradiation doses.

Conclusions:

• The proposed OCT and deep learning-based method provides an effective noninvasive approach for assessing laser-induced skin damage.
• HR-Net is a highly accurate model for segmenting and quantifying laser-induced skin damage volume.
• The study establishes a quantitative relationship between laser irradiation dose and skin damage volume.