Head and Neck Tumor Segmentation Using Pre-RT MRI Scans and Cascaded DualUNet.

Area of Science:

• Medical Imaging
• Radiotherapy
• Artificial Intelligence

Background:

• Accurate segmentation of primary gross tumor volumes and metastatic lymph nodes in head and neck cancer is critical for radiotherapy planning.
• High interobserver variability in manual tumor delineation necessitates effective automated segmentation tools.

Purpose of the Study:

• To develop and evaluate a dual-stage 3D UNet auto-segmentation approach for head and neck cancer pre-radiotherapy (pre-RT) MRI scans.
• To improve the accuracy and consistency of tumor delineation for radiotherapy treatment planning.

Main Methods:

• A dual-stage 3D UNet architecture using cascaded neural networks for progressive accuracy refinement was proposed.
• The method employed an ensemble of first-stage coarse models and second-stage refinement models for multiclass segmentation.
• The approach was trained and evaluated on the Head and Neck Tumor Segmentation for MR-Guided Applications (HNTS-MRG) 2024 Task 1 dataset, utilizing pre-RT and mid-RT T2-weighted MRI scans with 5-fold cross-validation.

Main Results:

• The dual-stage approach achieved a mean aggregated Dice similarity coefficient of 0.737 on the test set.
• Consistent performance improvements were observed across all cross-validation folds compared to a single-stage segmentation method.
• The ensemble of five coarse and ten refinement models demonstrated robust segmentation capabilities.

Conclusions:

• The proposed dual-stage 3D UNet approach offers a significant advancement in automated tumor segmentation for head and neck cancer radiotherapy.
• This method effectively refines segmentation accuracy, addressing the limitations of manual delineation and interobserver variability.
• The findings support the utility of advanced deep learning techniques for enhancing radiotherapy precision and patient outcomes.