Complex wound analysis using AI.

Area of Science:

• Biomedical Engineering
• Computational Pathology
• Wound Healing Research

Background:

• Impaired wound healing presents a significant clinical challenge.
• Current histological wound assessments are subjective, time-consuming, and often omitted.
• Macroscopic and limited histological methods lack comprehensive analysis of wound biopsies.

Purpose of the Study:

• To develop an automated deep learning (DL) approach for objective and comprehensive analysis of histological wound images.
• To apply this DL model to analyze complex human wounds and mouse models of wound healing.
• To improve the accuracy and efficiency of wound feature segmentation and quantification.

Main Methods:

• Development of a deep neural network (DNN) architecture optimized for segmenting characteristic wound features in H&E-stained images.
• Training and validation of the DL model using mouse wound biopsy images across four healing timepoints.
• Revision and application of the DL model for cellular-level analysis of human complex wound biopsies.

Main Results:

• The DL model achieved 89% mean test set accuracy in analyzing mouse wound healing.
• Model performance improved to 97% mean test set accuracy when analyzing human complex wounds at a cellular level.
• The approach enables comprehensive analysis of human wound biopsies and accurate morphometric analysis of mouse wound healing.

Conclusions:

• The automated DL approach provides objective and comprehensive analysis of H&E-stained wound sections.
• This method facilitates in-depth analysis of mouse wound healing and accurate morphometric quantification.
• The DL model aids in the analysis and quantification of immune cell infiltration for clinical diagnosis of human complex wounds.