Related Experiment Video
Updated: Oct 3, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
3.0K
Stepwise deep neural network (stepwise-net) for head and neck auto-segmentation on CT images
Daisuke Kawahara1, Masato Tsuneda2, Shuichi Ozawa3
1Department of Radiation Oncology, Graduate School of Biomedical Health Sciences, Hiroshima University, Hiroshima, 734-8551, Japan.
Computers in Biology and Medicine
|February 15, 2022
Summary
A new stepwise neural network (stepwise-net) improves auto-segmentation for head and neck cancer CT scans. This deep learning model outperforms traditional methods, enhancing radiotherapy treatment planning.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Radiotherapy
Background:
- Accurate segmentation of organs at risk in head and neck cancer is crucial for radiotherapy planning.
- Current auto-segmentation methods, including atlas-based approaches and conventional U-net models, have limitations in precision.
Purpose of the Study:
- To propose and evaluate a novel stepwise deep neural network (stepwise-net) for auto-segmentation of normal tissue structures in head and neck CT images.
- To compare the performance of the stepwise-net against a conventional U-net and an atlas-based segmentation method.
Main Methods:
- A stepwise neural network (stepwise-net) based on 3D Fully Convolutional Networks (FCN) was developed.
- The stepwise-net utilizes a two-stage approach: initial low-resolution identification of target regions, followed by high-resolution segmentation of cropped areas.
- Segmentation of brainstem, optic nerves, and parotid and submandibular glands was performed on 3D CT images.
Main Results:
- The stepwise-net achieved significantly higher Dice Similarity Coefficients (DSCs) and Jaccard Similarity Coefficients (JSCs) compared to the atlas-based method for all segmented organs.
- The stepwise-net demonstrated a significantly smaller Hausdorff distance (HD) than the atlas-based method.
- Compared to the conventional U-net, the stepwise-net exhibited superior DSC and JSC, with a smaller HD.
Conclusions:
- The proposed stepwise-network is superior to conventional U-net-based and atlas-based segmentation techniques.
- This deep learning model offers a potentially valuable method for enhancing the efficiency and accuracy of head and neck radiotherapy treatment planning.
