Skip to content
DraconiticResearch
Search Sources API
197,693 studies31 sourcesUpdated 43 min ago
Peer-reviewedCC BY 4.0Frontiers in Oncology

The feasibility of training nnU-Net to segment pediatric Wilms tumor with small datasets

  • M.A.D. Buser1
  • R.F. Meijer1
  • A.F.W. van der Steeg1
  • M.H.W.A. Wijnen1
  • M. M. van den Heuvel-Eibrink1,2
  • A.S. Littooij1,3
  • G. De Donno4
  • A. De Luca4
4 affiliations
  1. 1Princess Máxima Center for Pediatric Oncology
  2. 2Division of Childhealth, Wilhelmina Childrens’ Hospital
  3. 3Department of Radiology and Nuclear Medicine, University Medical Center Utrecht
  4. 4Image Sciences Institute, University Medical Center Utrecht, Utrecht University
Open at sourcePDFDOICite

Abstract

Introduction: Segmentation of magnetic resonance imaging (MRI) can be used in the clinical care of patients with (suspected) Wilms tumor (WT). Manual segmentation, however, is time consuming and user dependent. Deep learning can be used to automate this, but is often associated with the need for large training datasets to be trained successfully, which is inherently challenging in pediatric populations. The aim of this study was to determine the effect of the number of training patients on automated WT segmentation.

Methods: 56 patients with a suspected WT were included in this retrospective study. The pre-operative MRI scans were divided in a training (n = 45) and test (n = 11) set. The segmentation was based on nnU-Net, an out-of-the-box framework for image segmentation. nnU-Net was trained on a random selection of five patients, and the resulting segmentation method was tested on a fixed test set of 11 patients. Next, patients were added to the training dataset with incremental steps of five patients, up to the full dataset of 45 patients. This process was repeated 5 times to account for random sampling. The Dice score was used as measure for segmentation performance. We pragmatically defined a Dice score of > 0.80 as a reference level for segmentation performance.

Results: The lowest median Dice score was obtained when using a subset of five training patients (0.38). The highest median Dice score was observed for training on 45 patients (0.91). An median Dice score of 0.80 was reached using a minimal subset of 10 training patients. The Dice score showed dependency on the random selection of the training patients, with the largest spread in performance observed in the training set of five patients. A large spread in performance between the 11 test set patients was observed.

Discussion: Using this specific training and test cohort, a median Dice score of 0.80 was reached with just 10 training patients, however, the results showed considerable variation in the individual test patients. The performance spread between the random subsampling of the training patients highlights the importance of clinically representable datasets when dealing with the small dataset sizes inherent to pediatric oncology.

Related

  1. A prospective cross-sectional diagnostic feasibility study: musculoskeletal ultrasound for pediatric bone age evaluation
    Xingxu Zhou, Yifei Huang, Chunrong Liu +7
    Translational pediatricsPeer-reviewedCC BY-NC-NDFull textGeneral medicineEurope PMC
  2. Feasibility, Acceptability, and Initial Efficacy of a Wheelchair Exercise Training Program in People With Multiple Sclerosis
    Stephanie L. Silveira, Angela J. Piasecki, Robert W. Motl +1
    International Journal of MS CarePeer-reviewedLicence not statedFull textGeneral medicinePMC
  3. Training strategy over architecture: A systematic ablation study for Spartina detection from aerial imagery with small geospatial datasets
    Adrien Le Guillou, Manon Brehier, Jérôme Ammann +2
    PLoS ONEPeer-reviewedCC BY 4.0Full textStatisticsEndocrinology & metabolismHALPLOS
  4. Improving Awareness of Oman Medical Specialty Board Training Bylaws Among First-Year Pediatric Residents Using a Flipped Classroom Intervention: A Quality Improvement Project
    Saniya Al Husaini, Abdul Hamid Al Hinai, Majan Al Siyabi +2
    CureusPeer-reviewedCC BYFull textGeneral medicineEurope PMC
  5. Case Report: Multimodal treatment of pediatric H3G34-mutant diffuse hemispheric glioma: a safety and feasibility case report combining precision medicine and tumor treating fields with conventional care
    Toon Van Genechten, Alexander Claviez, Sebastian Oberbeck +11
    Frontiers in OncologyPeer-reviewedCC BY 4.0Full textOncologyFrontiersCrossrefDOAJ

Cite

BibTeXRIS
BibTeX
@article{buser2026feasibility,
  title = {{The feasibility of training nnU-Net to segment pediatric Wilms tumor with small datasets}},
  author = {M.A.D. Buser and R.F. Meijer and A.F.W. van der Steeg and M.H.W.A. Wijnen and M. M. van den Heuvel-Eibrink and A.S. Littooij and G. De Donno and A. De Luca},
  journal = {Frontiers in Oncology},
  year = {2026},
  month = oct,
  doi = {10.3389/fonc.2026.1841994},
  url = {https://www.frontiersin.org/articles/10.3389/fonc.2026.1841994/full},
  note = {CC BY 4.0},
}
RIS
TY  - JOUR
TI  - The feasibility of training nnU-Net to segment pediatric Wilms tumor with small datasets
AU  - M.A.D. Buser
AU  - R.F. Meijer
AU  - A.F.W. van der Steeg
AU  - M.H.W.A. Wijnen
AU  - M. M. van den Heuvel-Eibrink
AU  - A.S. Littooij
AU  - G. De Donno
AU  - A. De Luca
PY  - 2026
DA  - 2026/10/22
JO  - Frontiers in Oncology
DO  - 10.3389/fonc.2026.1841994
UR  - https://www.frontiersin.org/articles/10.3389/fonc.2026.1841994/full
AB  - Introduction: Segmentation of magnetic resonance imaging (MRI) can be used in the clinical care of patients with (suspected) Wilms tumor (WT). Manual segmentation, however, is time consuming and user dependent. Deep learning can be used to automate this, but is often associated with the need for large training datasets to be trained successfully, which is inherently challenging in pediatric populations. The aim of this study was to determine the effect of the number of training patients on automated WT segmentation.  Methods: 56 patients with a suspected WT were included in this retrospective study. The pre-operative MRI scans were divided in a training (n = 45) and test (n = 11) set. The segmentation was based on nnU-Net, an out-of-the-box framework for image segmentation. nnU-Net was trained on a random selection of five patients, and the resulting segmentation method was tested on a fixed test set of 11 patients. Next, patients were added to the training dataset with incremental steps of five patients, up to the full dataset of 45 patients. This process was repeated 5 times to account for random sampling. The Dice score was used as measure for segmentation performance. We pragmatically defined a Dice score of > 0.80 as a reference level for segmentation performance.  Results: The lowest median Dice score was obtained when using a subset of five training patients (0.38). The highest median Dice score was observed for training on 45 patients (0.91). An median Dice score of 0.80 was reached using a minimal subset of 10 training patients. The Dice score showed dependency on the random selection of the training patients, with the largest spread in performance observed in the training set of five patients. A large spread in performance between the 11 test set patients was observed.  Discussion: Using this specific training and test cohort, a median Dice score of 0.80 was reached with just 10 training patients, however, the results showed considerable variation in the individual test patients. The performance spread between the random subsampling of the training patients highlights the importance of clinically representable datasets when dealing with the small dataset sizes inherent to pediatric oncology.
SN  - 2234-943X
ER  -