Szczegóły publikacji
Opis bibliograficzny
Wound area detection with nnU-Net v2 architecture / Franciszek Kubala, Piotr Matusiewicz, Dominik Mika, Dominik Zawlocki, Zbisław TABOR // W: Joint 20th Nordic-Baltic Conference on Biomedical Engineering & 24th Polish Conference on Biocybernetics and Biomedical Engineering : joint Proceedings of NBC 2025 and PCBBE 2025, June 16–18, 2025, Warsaw, Poland / eds. Piotr Ładyżyński, Dorota G. Pijanowska, Adam Liebert. — Cham : Springer, cop. 2025. — (IFMBE Proceedings / International Federation for Medical & Biological Engineering ; ISSN 1680-0737 ; vol. 131). — ISBN: 978-3-031-96537-1; e-ISBN: 978-3-031-96538-8. — S. 468–476. — Bibliogr., Abstr.
Autorzy (5)
Słowa kluczowe
Dane bibliometryczne
| ID BaDAP | 161259 |
|---|---|
| Data dodania do BaDAP | 2025-07-25 |
| DOI | 10.1007/978-3-031-96538-8_40 |
| Rok publikacji | 2025 |
| Typ publikacji | materiały konferencyjne (aut.) |
| Otwarty dostęp |  |
| Wydawca | Springer |
| Czasopismo/seria | IFMBE Proceedings |
Abstract
Diabetic foot ulcers are a dangerous complication of diabetes and, if uncared for, can lead to significant health risks. In treatment, wound annotation and measurements are crucial. The development of AI models has the potential to significantly transform medical diagnostic techniques and greatly benefit the daily work of medical staff. With the support of deep learning, we aim to expand on new ways to implement convolutional neural networks (CNNs) in the daily work of medical staff. In this study we used the well renowned nnU-Net 2.0 architecture. The models are trained using custom loss function which improves segmentation accuracy compared to the state of the art approaches. Segmentation was performed using an ensemble of models. For the dataset, we obtained 451 wound pictures with ground truth segmentations of wounds and scale markers and split the dataset into a holdout test part of 15% images and 85% of training/validation images. Various models achieved accuracy of 90,91%, 90.88% and 91.65%. Model ended up with big accuracy and could be used in a medical setting.
