EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Detecting Potential Wound Healing Complications in Images Using Different Deep Learning Architectures

Dorothee A. Busch, Naga Venkata Sai Jitin Jami, Heike Leutheuser, Carola Berking, Cornelia Erfurt‐Berge, Ursula H. Hübner

International Wound Journal · 2026-11 · 23 (11)

DOI: 10.1111/iwj.71112 ↗

Kurz gefasst

Fachartikel in der Zeitschrift International Wound Journal, online erschienen am 19.11.2026. Verfasst von Dorothee A. Busch und 5 weiteren Autor:innen. Die Arbeit nennt 31 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Redness, oedema, maceration, necrosis and insufficient granulation are common signs of complications of chronic wounds. Early detection of these features in routine wound images is essential for timely, targeted treatment across care settings. Automated image analysis can standardize wound assessment and enable early recognition of complications, especially where specialist expertise is scarce. A new, clinical dataset of 1000 annotated images (200 per feature; balanced presence/absence) from patients with different types of chronic leg ulcers was compiled. Five convolutional neural networks of different sizes—MobileNetV2, DenseNet121, InceptionV3, ResNet50 and Xception—were trained and evaluated using primarily accuracy and AUC. Gradient‐weighted Class Activation Mapping was used to visualize the image regions relevant to the model predictions. Necrosis and lack of granulation achieved the highest performance (accuracy 0.9 and 0.8, AUC 0.98 and 0.96), while maceration and redness both reached an accuracy of 0.75 and an AUC of 0.89, and oedema was the most difficult to classify with an accuracy of 0.70 and an AUC of 0.64. Deep learning can accurately identify complications in wound healing from routine images with necrosis and lack of granulation classified most reliably, although classification performance varies across complication types.

Zitieren

APA

Busch, D. A., Jami, N. V. S. J., Leutheuser, H., Berking, C., Erfurt‐Berge, C., & Hübner, U. H. (2026). Detecting Potential Wound Healing Complications in Images Using Different Deep Learning Architectures. International Wound Journal, 23(11). https://doi.org/10.1111/iwj.71112

Vancouver

Busch DA, Jami NVSJ, Leutheuser H, Berking C, Erfurt‐Berge C, Hübner UH. Detecting Potential Wound Healing Complications in Images Using Different Deep Learning Architectures. International Wound Journal. 2026;23(11). doi:10.1111/iwj.71112

BibTeX
@article{Busch2026,
  author = {Busch, Dorothee A. and Jami, Naga Venkata Sai Jitin and Leutheuser, Heike and Berking, Carola and Erfurt‐Berge, Cornelia and Hübner, Ursula H.},
  title = {Detecting Potential Wound Healing Complications in Images Using Different Deep Learning Architectures},
  journal = {International Wound Journal},
  year = {2026},
  volume = {23},
  number = {11},
  doi = {10.1111/iwj.71112}
}
Kostenlos starten →