Related Experiment Video
Updated: May 9, 2025
10:48
PET and MRI Guided Irradiation of a Glioblastoma Rat Model Using a Micro-irradiator
Published on: December 28, 2017
9.4K
Deep Learning for Longitudinal Gross Tumor Volume Segmentation in MRI-Guided Adaptive Radiotherapy for Head and Neck
Xin Tie1, Weijie Chen1, Zachary Huemann1
1University of Wisconsin, Madison, WI, USA.
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
|April 29, 2025
Summary
Deep learning models accurately segmented gross tumor volumes (GTV) for head and neck cancer radiotherapy. Combining pre- and mid-treatment data improved segmentation accuracy, streamlining workflows.
Area of Science:
- Medical Imaging
- Radiotherapy
- Artificial Intelligence
Background:
- Accurate gross tumor volume (GTV) segmentation is critical for MRI-guided adaptive radiotherapy (MRgART) in head and neck cancer.
- Manual segmentation is time-consuming and suffers from interobserver variability.
- Deep learning (DL) offers automated GTV delineation to address these limitations.
Purpose of the Study:
- To develop and evaluate DL models for longitudinal GTV segmentation in MRgART for head and neck cancer.
- To tackle segmentation challenges for both pre-radiotherapy (pre-RT) and mid-radiotherapy (mid-RT) phases.
- To improve the accuracy and efficiency of GTV segmentation in radiation oncology workflows.
Main Methods:
- Developed DL models using SegResNet with deep supervision as the backbone.
- Task 1 (pre-RT): Trained models on combined pre-RT and mid-RT MRI data.
- Task 2 (mid-RT): Introduced mask-aware attention modules for incorporating pre-RT information into mid-RT segmentation.
Main Results:
- Combined pre-RT and mid-RT data improved aggregated Dice similarity coefficient (DSC_agg) for Task 1 compared to pre-RT data alone.
- Attention-based approach in Task 2 showed slight improvements over baseline concatenation.
- Ensemble models achieved 1st place, with average DSC_agg of 0.794 (Task 1) and 0.733 (Task 2).
Conclusions:
- The developed DL models effectively facilitate GTV segmentation in MRgART.
- Longitudinal data integration and attention mechanisms enhance segmentation performance.
- These models have the potential to streamline radiation oncology workflows and improve treatment delivery.
