Gradient Map-Assisted Head and Neck Tumor Segmentation: A Pre-RT to Mid-RT Approach in MRI-Guided Radiotherapy

TL;DR

The paper addresses accurate GTV segmentation in MRI-guided adaptive radiotherapy for head and neck cancer by leveraging pre-RT tumor locations as priors and gradient maps computed within deformably registered bounding boxes to inform mid-RT segmentation. An nnUNet-based model processes two-channel inputs consisting of mid-RT T2w images and gradient maps, with data augmentation via independent pre-RT and mid-RT samples. Across 5-fold cross-validation and a test submission, the approach improves segmentation, achieving $DSC_{agg}$ values of $0.534$ for $GTVp$ and $0.867$ for $GTVn$ (mean $0.70$), with larger gains for $GTVn$ and more modest improvements for $GTVp$ due to tumor shrinkage and boundary ambiguity. The study highlights the potential of temporal priors and local gradient information to enhance adaptive radiotherapy planning, while pointing to data limitations and the promise of human-in-the-loop or probabilistic uncertainty models to further boost performance. Overall, the method offers a practical pathway to incorporate pre-RT temporal information into mid-RT segmentation for MRI-guided HNC treatment.

Abstract

Radiation therapy (RT) is a vital part of treatment for head and neck cancer, where accurate segmentation of gross tumor volume (GTV) is essential for effective treatment planning. This study investigates the use of pre-RT tumor regions and local gradient maps to enhance mid-RT tumor segmentation for head and neck cancer in MRI-guided adaptive radiotherapy. By leveraging pre-RT images and their segmentations as prior knowledge, we address the challenge of tumor localization in mid-RT segmentation. A gradient map of the tumor region from the pre-RT image is computed and applied to mid-RT images to improve tumor boundary delineation. Our approach demonstrated improved segmentation accuracy for both primary GTV (GTVp) and nodal GTV (GTVn), though performance was limited by data constraints. The final DSCagg scores from the challenge's test set evaluation were 0.534 for GTVp, 0.867 for GTVn, and a mean score of 0.70. This method shows potential for enhancing segmentation and treatment planning in adaptive radiotherapy. Team: DCPT-Stine's group.

Figures (4)

• Figure 1: Workflow for incorporating pre-RT tumor location and gradient maps in mid-RT segmentation.
• Figure 2: Comparison of segmentation performance with (w/) and without (w/o) gradient maps on fold-0 of the validation set (n=30).
• Figure 3: Example cases demonstrating the impact of using gradient maps. Patient a: The gradient map enabled accurate segmentation of a previously undetected GTVn, improving the DSC from 0.0 to 0.83. Patient b: The use of the gradient map increased the DSC for both GTVp (from 0.14 to 0.80) and GTVn (from 0.73 to 0.87), although part of the GTVp was missed as it extended beyond the gradient map's bounding box.
• Figure 4: Relationship between tumor volume change and DSC scores for GTVp and GTVn. (a) The first row shows three subplots representing different bins of mid-RT GTVp volumes. Each subplot's x-axis represents the change in volume (pre-RT volume minus mid-RT volume), and the y-axis represents the predicted DSC score. The size of the dots corresponds to the mid-RT GTV volume, while red X marks indicate instances where the current mid-RT volume is 0. (b) The second row displays similar subplots for GTVn, illustrating the relationship between volume change and segmentation accuracy.