An innovative machine learning approach to wound edge assessment using depth maps and wound border rectification

Chronic wounds represent a widespread and costly healthcare crisis, undermining patients' quality of life and challenging clinicians with subjective and inconsistent assessment methods. This thesis aims to enhance wound edge assessment by introducing a innovative method based both on advanced Artificial Intelligence models and cutting-edge image processing techniques. A unique feature of the method is the integration of wound depth maps with a novel wound edge rectification technique. A dataset of 159 annotated wound images, provided by IRCCS Sant'Orsola-Malpighi University Hospital, was analyzed. Depth maps were used to capture the three-dimensional structural details of wound edges. The wound border rectification method was applied to standardize the periwound area geometries, enabling consistent depth feature extraction. Clustering algorithms identified natural groupings of wound edge features in agreement with of BWAT scoring. Moreover, it identified 8 clusters providing insights into underlying structural patterns to be investigated. A Random Forest model based on extracted depth features was employed., encountering similar challenges as clinicians in differentiating between ambiguous wound edge types; however, by merging these categories, it achieved a strong accordance with clinician. The results validate the potential of Machine Learning approaches to improve diagnostic accuracy. The clustering analysis highlights new avenues for exploring the micro-structural organization of wound edges, paving the way for more detailed investigations. The proposed method offers a fully automated solution for image characterization of wound edges, aiming to overcome the limitations of traditional manual assessments and to support clinicians with objective, consistent, and reproducible insights. This work lays the foundation for future advancements in automated wound care systems, with the goal of optimizing patient outcomes and healthcare resource utilization.