Related Experiment Video
Updated: Jul 23, 2025
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
Deep Learning Architecture to Improve Edge Accuracy of Auto-Contouring for Head and Neck Radiotherapy
Ryan Gifford1, Sachin R Jhawar2, Samantha Krening1
1Department of Integrated Systems Engineering, The Ohio State University, 1971 Neil Ave, Columbus, OH 43210, USA.
Abstract:
Deep learning (DL) methods have shown great promise in auto-segmentation problems. However, for head and neck cancer, we show that DL methods fail at the axial edges of the gross tumor volume (GTV) where the segmentation is dependent on information closer to the center of the tumor. These failures may decrease trust and usage of proposed auto-contouring systems. To increase performance at the axial edges, we propose the spatially adjusted recurrent convolution U-Net (SARC U-Net). Our method uses convolutional recurrent neural networks and spatial transformer networks to push information from salient regions out to the axial edges. On average, our model increased the Sørensen-Dice coefficient (DSC) at the axial edges of the GTV by 11% inferiorly and 19.3% superiorly over a baseline 2D U-Net, which has no inherent way to capture information between adjacent slices. Over all slices, our proposed architecture achieved a DSC of 0.613, whereas a 3D and 2D U-Net achieved a DSC of 0.586 and 0.540, respectively. SARC U-Net can increase accuracy at the axial edges of GTV contours while also increasing accuracy over baseline models, creating a more robust contour.
Insights
A new deep learning model, SARC U-Net, improves head and neck cancer auto-segmentation, particularly at tumor edges. This enhances contour accuracy and reliability for clinical use.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Oncology
Background:
- Deep learning (DL) methods show promise in medical image auto-segmentation.
- Existing DL models struggle with accurate segmentation of head and neck tumors at axial edges.
- Segmentation failures can reduce trust and clinical adoption of auto-contouring systems.
Purpose of the Study:
- To develop an improved DL method for accurate head and neck tumor auto-segmentation.
- To address the specific challenge of segmenting the gross tumor volume (GTV) at its axial edges.
- To enhance the robustness and reliability of auto-contouring systems.
Main Methods:
- Proposed the spatially adjusted recurrent convolution U-Net (SARC U-Net).
- Utilized convolutional recurrent neural networks and spatial transformer networks to integrate information across slices.
- Focused on pushing information from salient regions to the axial edges for improved segmentation.
Main Results:
- SARC U-Net improved Sørensen-Dice coefficient (DSC) at GTV axial edges by 11% inferiorly and 19.3% superiorly compared to a baseline 2D U-Net.
- Achieved an overall DSC of 0.613, outperforming 3D U-Net (0.586) and 2D U-Net (0.540).
- Demonstrated increased accuracy at tumor edges and overall segmentation performance.
Conclusions:
- SARC U-Net enhances accuracy for head and neck GTV contours, especially at axial edges.
- The proposed method offers a more robust auto-contouring solution compared to baseline DL models.
- Improved segmentation accuracy can increase clinical trust and utility of AI in radiation oncology.
