Automated Extraction of Skin Wound Healing Biomarkers From In Vivo Label-Free Multiphoton Microscopy Using...

Area of Science:

• Biomedical imaging
• Computational pathology
• Wound healing research

Background:

• Histological analysis is standard for wound healing studies.
• Label-free multiphoton microscopy (MPM) offers natural contrast and metabolic insights.
• Manual segmentation of MPM images is time-consuming, limiting clinical application.

Observation:

• Two convolutional neural networks (CNNs) were trained for MPM image segmentation.
• One CNN segmented wound tissue sections (380 images), the other segmented in vivo wound edge z-stacks (5,848 images).
• Network accuracy was validated against manual measurements.

Findings:

• The wound section CNN achieved 92.83% accuracy, with measurements within 10% error.
• The in vivo CNN achieved 89.66% accuracy, quantifying optical redox ratio within 5% error.
• Automated segmentation enabled accurate wound geometry and metabolic quantification.

Implications:

• CNNs enable automated, rapid quantification of skin wound geometry and metabolism using MPM.
• This overcomes the limitations of manual segmentation, making MPM more practical for wound monitoring.
• MPM with CNNs can provide near real-time quantitative tissue structure and function readouts.