Head and neck tumor segmentation in PET/CT: The HECKTOR challenge
Valentin Oreiller1, Vincent Andrearczyk2, Mario Jreige3
1Institute of Information Systems, University of Applied Sciences Western Switzerland (HES-SO), Sierre, Switzerland; Department of Nuclear Medicine and Molecular Imaging, Centre Hospitalier Universitaire Vaudois (CHUV), Lausanne, Switzerland.
Medical Image Analysis
|January 11, 2022
Summary
The HECKTOR challenge advanced automatic segmentation of head and neck tumors using combined PET/CT scans. AI methods achieved superior accuracy (DSC 0.7591) compared to human segmentation, aiding future radiomics research.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Oncology
Background:
- The HEad and neCK TumOR (HECKTOR) challenge was the first to focus on segmenting Gross Tumor Volume (GTV) in combined FDG-PET/CT images for head and neck cancers.
- Manual segmentation of GTVs is time-consuming and prone to errors, hindering large-scale radiomics studies.
Purpose of the Study:
- To evaluate automated GTV segmentation methods for head and neck primary tumors using combined FDG-PET/CT.
- To compare the performance of automated methods against human inter-observer agreement and single-modality approaches.
Main Methods:
- The HECKTOR challenge provided a training dataset (201 cases) and a test dataset (53 cases) for evaluating segmentation algorithms.
- Methods were ranked using the Dice Score Coefficient (DSC) averaged across test cases.
- An inter-observer agreement study was conducted to establish a human performance benchmark.
Main Results:
- The best automated method achieved an average DSC of 0.7591, significantly outperforming the baseline method (DSC 0.6610) and human inter-observer agreement (DSC 0.61).
- Automated methods effectively utilized both metabolic (PET) and structural (CT) information.
- AI-driven segmentation surpassed single-modality and semi-automatic thresholding techniques.
Conclusions:
- Automated GTV segmentation in head and neck cancer using combined PET/CT shows high potential.
- These advanced methods offer a more efficient and accurate alternative to manual delineation, facilitating radiomics research.
- The HECKTOR challenge demonstrated the capability of AI to improve tumor segmentation accuracy in oncology.
