Abstract
Background/Objectives: This review systematically assesses machine learning (ML) and deep learning (DL) applications using images to diagnose diabetic foot ulcers (DFUs), focusing on detection, segmentation, and classification. The study explores trends, challenges, and quality measurements of the reviewed research. Methods: A comprehensive search was conducted in October 2025 across 14 databases, covering studies published between 2010 and 2025. Studies employing ML/DL for DFU diagnosis with accurate measurements were included, while those without image-based methods, AI techniques, or relevant outcomes were excluded. Out of 4653 articles initially identified, 1016 underwent detailed review, and 102 met the inclusion criteria. Results: The analysis revealed that ML/DL models are effective tools for DFU diagnosis, achieving accuracy between 0.88 and 0.97, specificity between 0.85 and 0.95, and sensitivity between 0.89 and 0.95. Common methods included Support Vector Machines (SVMs) for ML and U-Net or fully convolutional neural networks (FCNNs) for DL. Recent studies also explored thermal infrared imaging as a promising diagnostic technique. However, only 45% of segmentation datasets and 67.3% of classification datasets were publicly accessible, limiting reproducibility and further development. Conclusions: This review provides valuable insights into trends and key findings in ML/DL applications for DFU diagnosis. It highlights the need for improved data availability and sharing to enhance reproducibility, accuracy, and reliability, ultimately improving patient care.