Related Experiment Video
Updated: May 15, 2026
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
Robust automatic soft tissue flap segmentation using a challenging case-enriched nnU-Net in head and neck CT images
Abir Fathallah1, Zacharia Mesbah2,3, Alice Blache4
1LaTIM, INSERM, UMR 1101, Univ Brest, Brest, France.
Scientific Reports
|May 13, 2026
Summary
Automating flap segmentation in reconstructive surgery improves radiotherapy planning. Enhancing deep learning models with challenging cases significantly boosts segmentation accuracy and robustness for better clinical outcomes.
Area of Science:
- Medical imaging
- Radiotherapy planning
- Deep learning in medicine
Background:
- Reconstructive surgery flaps complicate radiotherapy volume definition and automatic segmentation of organs-at-risk and nodal volumes.
- Automated flap segmentation is crucial for accurate radiotherapy planning and monitoring flap changes post-treatment.
Purpose of the Study:
- To investigate if enriching a training dataset with challenging cases improves automated flap segmentation accuracy and robustness.
- To enhance the performance of the nnU-Net deep learning model for flap segmentation in radiotherapy.
Main Methods:
- Enriched a previously developed training dataset with challenging cases, including pedicled flaps, small flaps, maxillary flaps, bone resections, and dental artifacts.
- Utilized the nnU-Net deep learning architecture to train the model on the enriched dataset.
- Evaluated segmentation performance using Dice scores and compared results with paired Wilcoxon signed-rank tests.
Main Results:
- The enriched training dataset significantly improved the nnU-Net model's performance.
- Mean Dice scores increased from 0.66 ± 0.29 to 0.74 ± 0.20 (p < 0.001).
- Median Dice scores improved from 0.76 to 0.80, demonstrating enhanced robustness.
Conclusions:
- Enriching training datasets with diverse and challenging cases is an effective strategy to improve automated flap segmentation.
- The enhanced nnU-Net model provides robust flap segmentation without architectural modifications, aiding in radiotherapy planning and analysis of flap changes.