EUVIMEDEUVIMED entdecken
Crossref

G-T correcting: an improved training of image segmentation under noisy labels

Yun Gao, Junhu Fu, Yi Guo, Yuanyuan Wang

Medical & Biological Engineering & Computing · 2024-12 · 62 (12) · 3781-3799

DOI: 10.1007/s11517-024-03170-4 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Medical & Biological Engineering & Computing, online erschienen am 20.07.2024. Verfasst von Yun Gao und 3 weiteren Autor:innen. Beteiligte Einrichtung: School of Information Science and Technology of Fudan University, 220 Handan Rd, Shanghai, 200433, China. Hauptthemen laut MEDLINE: Image Processing, Computer-Assisted, Diagnostic Imaging, Signal-To-Noise Ratio. Die Arbeit wurde bisher 1-mal zitiert (Crossref) und nennt 33 Literaturangaben. Gefördert von: Science and Technology Commission of Shanghai Municipality.

AbstractSchlagwörterBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Data-driven medical image segmentation networks require expert annotations, which are hard to obtain. Non-expert annotations are often used instead, but these can be inaccurate (referred to as "noisy labels"), misleading the network's training and causing a decline in segmentation performance. In this study, we focus on improving the segmentation performance of neural networks when trained with noisy annotations. Specifically, we propose a two-stage framework named "G-T correcting," consisting of "G" stage for recognizing noisy labels and "T" stage for correcting noisy labels. In the "G" stage, a positive feedback method is proposed to automatically recognize noisy samples, using a Gaussian mixed model to classify clean and noisy labels through the per-sample loss histogram. In the "T" stage, a confident correcting strategy and early learning strategy are adopted to allow the segmentation network to receive productive guidance from noisy labels. Experiments on simulated and real-world noisy labels show that this method can achieve over 90% accuracy in recognizing noisy labels, and improve the network's DICE coefficient to 91%. The results demonstrate that the proposed method can enhance the segmentation performance of the network when trained with noisy labels, indicating good clinical application prospects.

Schlagwörter

MeSH-Schlagwörter (MEDLINE)

Image Processing, Computer-AssistedDiagnostic ImagingSignal-To-Noise RatioNormal DistributionReproducibility of ResultsNeural Networks, ComputerHumans

Stichwörter der Autor:innen

Label correctionLabel noise recognitionMedical image segmentationNoisy labels

Autor:innen und Institutionen

  • School of Information Science and Technology of Fudan University, 220 Handan Rd, Shanghai, 200433, China.
  • Key Laboratory of Medical Imaging Computing and Computer Assisted Intervention (MICCAI) of Shanghai, Shanghai, 200032, China.
  • School of Information Science and Technology of Fudan University, 220 Handan Rd, Shanghai, 200433, China. guoyi@fudan.edu.cn.
  • Key Laboratory of Medical Imaging Computing and Computer Assisted Intervention (MICCAI) of Shanghai, Shanghai, 200032, China. guoyi@fudan.edu.cn.
  • School of Information Science and Technology of Fudan University, 220 Handan Rd, Shanghai, 200433, China. yywang@fudan.edu.cn.
  • Key Laboratory of Medical Imaging Computing and Computer Assisted Intervention (MICCAI) of Shanghai, Shanghai, 200032, China. yywang@fudan.edu.cn.

Förderung

  • Science and Technology Commission of Shanghai Municipality

Zitieren

APA

Gao, Y., Fu, J., Guo, Y., & Wang, Y. (2024). G-T correcting: an improved training of image segmentation under noisy labels. Medical & Biological Engineering & Computing, 62(12), 3781-3799. https://doi.org/10.1007/s11517-024-03170-4

Vancouver

Gao Y, Fu J, Guo Y, Wang Y. G-T correcting: an improved training of image segmentation under noisy labels. Medical & Biological Engineering & Computing. 2024;62(12):3781-3799. doi:10.1007/s11517-024-03170-4

BibTeX
@article{Gao2024,
  author = {Gao, Yun and Fu, Junhu and Guo, Yi and Wang, Yuanyuan},
  title = {G-T correcting: an improved training of image segmentation under noisy labels},
  journal = {Medical & Biological Engineering & Computing},
  year = {2024},
  volume = {62},
  number = {12},
  pages = {3781--3799},
  doi = {10.1007/s11517-024-03170-4}
}
Kostenlos starten →