Related Experiment Video
Updated: Jun 15, 2025
08:41
Patient-Specific Polyvinyl Alcohol Phantom Fabrication with Ultrasound and X-Ray Contrast for Brain Tumor Surgery Planning
Published on: July 14, 2020
8.4K
Head and Neck Gross Tumor Volume Automatic Segmentation Using PocketNet
Awj Twam1, Adrian Celaya1, Evan Lim2
1Department of Imaging Physics, The University of Texas MD Anderson Cancer Center, Houston, TX, 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
|June 11, 2025
Summary
Team Pocket utilized PocketNet, a lightweight convolutional neural network (CNN), for automated head and neck cancer (HNC) tumor segmentation. This AI approach achieved promising results in segmenting primary and nodal tumors in MR images, aiding treatment planning.
Area of Science:
- Medical imaging and artificial intelligence
- Oncology and radiation therapy
- Computational anatomy
Background:
- Head and neck cancer (HNC) poses a significant global health challenge, necessitating precise tumor volume delineation for effective treatment planning.
- Manual segmentation of gross tumor volumes (GTV) in MR images is time-consuming, labor-intensive, and subject to inter-observer variability.
- Automated segmentation techniques, particularly those employing deep learning, are crucial for improving efficiency and consistency in HNC treatment.
Purpose of the Study:
- To evaluate the efficacy of PocketNet, a lightweight convolutional neural network (CNN), for automated segmentation of primary (GTVp) and nodal (GTVn) gross tumor volumes in head and neck cancer (HNC) from pre-radiotherapy MR images.
- To assess the performance of PocketNet within the context of the HNTS-MRG 2024 Grand Challenge, Task 1.
- To demonstrate the potential of automated segmentation for enhancing HNC treatment workflows.
Main Methods:
- Application of PocketNet, a lightweight CNN architecture, for the segmentation of GTVp and GTVn in MR images.
- Participation in Task 1 of the HNTS-MRG 2024 Grand Challenge, focusing on pre-radiotherapy MR data.
- Quantitative evaluation using the Dice Sorensen Coefficient (DSCagg) for performance assessment.
Main Results:
- PocketNet achieved an aggregated Dice Sorensen Coefficient (DSCagg) of 0.808 for GTVn and 0.732 for GTVp.
- The overall mean performance across both tumor types was 0.77.
- These results indicate robust performance in automated tumor segmentation for HNC.
Conclusions:
- PocketNet demonstrates significant potential as an efficient and accurate tool for automated GTV segmentation in MR-guided HNC interventions.
- The developed approach shows promise for integration into clinical workflows, potentially improving treatment planning and delivery.
- Further optimization of the PocketNet architecture and training strategies may lead to enhanced segmentation accuracy for HNC.
