Related Experiment Video
Updated: Aug 5, 2025
08:17
Dynamic Lung Tumor Tracking for Stereotactic Ablative Body Radiation Therapy
Published on: June 7, 2015
15.8K
Validation of a Fully Automated Hybrid Deep Learning Cardiac Substructure Segmentation Tool for Contouring and Dose
V Chin1, R N Finnegan2, P Chlap1
1University of New South Wales, South Western Sydney Clinical School, Sydney, Australia; Ingham Institute for Applied Medical Research, Radiation Oncology, Sydney, Australia; Department of Radiation Oncology, Liverpool and Macarthur Cancer Therapy Centres, Sydney, Australia.
Clinical Oncology (Royal College of Radiologists (Great Britain))
|March 24, 2023
Summary
A new hybrid tool accurately segments cardiac substructures for radiation dose evaluation. This method shows promise for large-scale studies on radiation-related heart damage.
Area of Science:
- Medical imaging analysis
- Radiation oncology
- Cardiovascular research
Background:
- Accurate delineation of cardiac substructures is crucial but challenging.
- Variations in anatomy and imaging present difficulties for segmentation.
- Radiation therapy requires precise cardiac substructure segmentation for dose evaluation.
Purpose of the Study:
- To validate a novel hybrid automatic segmentation tool for cardiac structures.
- To assess the tool's accuracy and consistency in dose evaluation.
- To explore its potential in clinical settings and cardiotoxicity studies.
Main Methods:
- A hybrid method combining deep learning, multi-atlas, and geometric segmentation was used.
- The tool automatically segmented 18 cardiac structures.
- Validation was performed on 30 lung cancer cases with diverse imaging variations.
Main Results:
- The tool achieved high accuracy, with median Dice Similarity Coefficients (DSC) of 0.75-0.93.
- Median Mean Distance to Agreement (MDA) ranged from 2.09-3.34 mm for major structures.
- Dose differences for cardiac substructures were within acceptable limits.
Conclusions:
- The novel hybrid automatic segmentation tool demonstrates high accuracy and consistency.
- It is suitable for challenging anatomical and imaging variations.
- The tool has promising applications in large-scale cardiotoxicity studies and dose calculations.
