Related Experiment Video
Updated: Aug 22, 2025
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
Patient-specific daily updated deep learning auto-segmentation for MRI-guided adaptive radiotherapy
Zhenjiang Li1, Wei Zhang2, Baosheng Li1
1Department of Radiation Oncology Physics and Technology, Shandong Cancer Hospital and Institute, Shandong First Medical University and Shandong Academy of Medical Sciences, No.440, Jiyan Road, Jinan 250117, Shandong Province, P.R.China.
This study introduces a patient-specific deep learning auto-segmentation strategy that enhances contouring accuracy and efficiency in MR-guided adaptive radiotherapy (MRgART). The method significantly reduces contouring time, facilitating routine clinical practice.
Area of Science:
- Medical Imaging
- Radiotherapy
- Artificial Intelligence
Background:
- Deep learning (DL) shows promise for MR-guided adaptive radiotherapy (MRgART) but struggles with online contouring accuracy.
- Current methods often lack patient-specific adaptation, limiting performance in longitudinal treatments.
Purpose of the Study:
- To develop and evaluate a patient-specific DL auto-segmentation (DLAS) strategy for improving online contouring in MRgART.
- To enhance segmentation accuracy and efficiency by leveraging previous patient data for model updates.
Main Methods:
- A patient-specific DL model was trained using initial MRI scans and contours.
- The model was iteratively updated with data from subsequent fractions, incorporating consistency constraints.
- Performance was assessed using Dice Similarity Coefficient (DSC) and 95% Hausdorff Distance (HD95), comparing against deformable image registration (DIR) and a non-updating DL model.
Main Results:
- The patient-specific DLAS achieved superior segmentation accuracy, with a mean DSC of 0.90 compared to 0.63 for DIR and 0.74 for the frozen DL model.
- Tumor segmentation yielded a median DSC of 0.95 and a median HD95 of 1.63 mm.
- Contouring time was significantly reduced from manual processes (12-22 mins) to 73.4 seconds, and online adaptive radiotherapy (ART) time decreased substantially.
Conclusions:
- The proposed patient-specific DLAS method significantly improves segmentation accuracy and efficiency for longitudinal MRIs in MRgART.
- This approach facilitates the routine clinical adoption of MR-guided adaptive radiotherapy.
