University of Twente Student Theses
Automatic Delineation of Laryngeal and Hypopharyngeal Gross Tumor Volume: A Pathologically and Clinically Validated Deep Learning Model
Kuijer, K.M. (2023) Automatic Delineation of Laryngeal and Hypopharyngeal Gross Tumor Volume: A Pathologically and Clinically Validated Deep Learning Model.
Full text not available from this repository.
| Full Text Status: | Access to this publication is restricted |
| Abstract: | Radiotherapy is essential for laryngeal and hypopharyngeal cancer treatment, requiring accurate gross tumor volume (GTV) delineation for optimal outcomes. Manual GTV delineation is time-consuming, prone to variability, and often overestimates tumor volume. This study aimed to develop a deep learning model for automated GTV delineation and evaluate its clinical utility. Two 3D convolutional neural networks (nnUNet and DeepMedic) were trained on MRI-only and MRI-PET-CT imaging inputs using a dataset of 193 patients. A dataset of 18 patients with pathologically delineated tumors was used for additional evaluation. Quantitative evaluation against manual and pathological delineations utilized Dice similarity coefficient (DSC). Additionally, acceptability for treatment planning was qualitatively assessed and a semi-automatic workflow was evaluated. Results showed median DSC between automatic and manual delineations of 0.58, 0.60 for MRI-only and 0.79, 0.72 for MRI-PET-CT input for nnUNet and DeepMedic, respectively. The nnUNet MRI-only delineations were deemed clinically acceptable in 67% of cases, versus 63% for manual and 40% for DeepMedic delineations. Semi-automatic corrections achieved a median general conformity index of 0.72 (MRI-only) and 0.68 (MRI-PET-CT). Our deep learning models achieved comparable accuracy to manual delineations, with potential for improved consistency and efficiency in treatment planning workflows. |
| Item Type: | Essay (Master) |
| Faculty: | TNW: Science and Technology |
| Subject: | 44 medicine |
| Programme: | Technical Medicine MSc (60033) |
| Link to this item: | https://purl.utwente.nl/essays/96937 |
| Export this item as: | BibTeX EndNote HTML Citation Reference Manager |
Show download statistics for this publication
Show download statistics for this publicationRepository Staff Only: item control page