Related Experiment Video
Updated: Sep 17, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
3.0K
Deep learning for automated, motion-resolved tumor segmentation in radiotherapy
Sagnik Sarkar1, P Troy Teo1, Mohamed E Abazeed2,3
1Department of Radiation Oncology, Feinberg School of Medicine, Northwestern University, Chicago, IL, USA.
NPJ Precision Oncology
|June 30, 2025
Summary
A new deep learning model, iSeg, automates lung tumor segmentation for radiotherapy, creating internal target volumes (ITVs) that capture motion. This AI approach matches human accuracy and improves precision, potentially enhancing treatment outcomes.
Area of Science:
- Medical Physics
- Radiotherapy
- Artificial Intelligence in Oncology
Background:
- Accurate tumor delineation is critical for effective radiotherapy.
- Manual segmentation is time-consuming and prone to variability.
- Deep learning offers potential for automating this process.
Purpose of the Study:
- To develop and validate a deep neural network (iSeg) for segmenting lung tumors and generating internal target volumes (ITVs).
- To assess the model's performance across multiple institutions and its robustness to image quality and tumor motion.
Main Methods:
- A 3D UNet deep learning model (iSeg) was trained on 739 CT scans from a multicenter registry.
- The model segmented gross tumor volumes (GTVs) and propagated them across 4D CT images to create ITVs.
- Validation was performed on two independent external cohorts (n=161 and n=102).
Main Results:
- iSeg achieved a median Dice Similarity Coefficient (DSC) of 0.73 in the internal cohort and comparable performance in external validation (DSC=0.70, 0.71).
- The model's performance matched human inter-observer variability and demonstrated robustness to image quality and motion.
- Machine-generated ITVs were significantly smaller than physician-delineated contours (p<0.0001), indicating increased precision.
Conclusions:
- The iSeg model represents a significant advancement in automated target volume segmentation for radiotherapy.
- Machine delineation enhances accuracy, reproducibility, and efficiency in radiotherapy planning.
- False positive rates in machine segmentation may correlate with clinical outcomes, warranting further investigation.
