Researchers describe an explainable AI system for outlining diabetic foot ulcers from clinical photographs, emphasizing attribution for where the model “looked” when making decisions. The work, published in BMC Medical Imaging, used a hybrid architecture combining a vision transformer bottleneck with convolutional localization (TransUNet-GradCAM) for pixel-accurate segmentation. Because diabetic foot outcomes depend on tracking progression over time, automated, interpretable segmentation could help standardize wound assessments that often rely on subjective manual tracing. The explainability component aims to improve clinician trust by tying model outputs to image regions. For biotech in digital health, the study adds another example of explainable medical imaging approaches moving toward clinical validation, though real-world performance and generalization beyond the training dataset remain critical next steps.