Related Experiment Video
Updated: May 23, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.6K
Assessing Self-supervised xLSTM-UNet Architectures for Head and Neck Tumor Segmentation in MR-Guided Applications
Abdul Qayyum1, Moona Mazher2, Steven A Niederer1
1National Heart and Lung Institute, Faculty of Medicine, Imperial College London, London, UK.
Head and Neck Tumor Segmentation for Mr-Guided Applications : First MICCAI Challenge, HNTS-MRG 2024, Held in Conjunction with MICCAI 2024, Marrakesh, Morocco, October 17, 2024, Proceedings
|May 22, 2025
Summary
A new two-stage model enhances head and neck tumor segmentation for MRI-guided radiation therapy (RT). This approach improves precision and minimizes side effects in cancer treatment.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Oncology
Background:
- Radiation therapy (RT) is crucial for head and neck cancer (HNC).
- MRI-guided RT offers enhanced precision and reduced side effects.
- Accurate tumor segmentation is vital for adaptive RT planning.
Purpose of the Study:
- To develop a novel two-stage model for Head and Neck Tumor Segmentation (HNTS) optimized for MRI-guided adaptive RT.
- To address data scarcity challenges in medical imaging using self-supervised learning.
- To improve segmentation accuracy by integrating spatial and temporal features.
Main Methods:
- Utilized a Self-Supervised 3D Student-Teacher Learning Framework with DINOv2 for representation learning on unlabeled data.
- Fine-tuned an xLSTM-based UNet model to capture spatial and sequential tumor features.
- Evaluated the model on diverse HNC cases for automated segmentation.
Main Results:
- Achieved a mean aggregated Dice Coefficient of 0.81 for pre-RT segments.
- Achieved a mean aggregated Dice Coefficient of 0.65 for mid-RT segments.
- Demonstrated significant improvements over state-of-the-art deep learning models.
Conclusions:
- The proposed two-stage model provides a robust and generalizable solution for automated HNTS.
- This advancement enhances MRI-guided adaptive RT planning for HNC patients.
- The work contributes to improved quality of care in HNC treatment.
