Related Experiment Video
Updated: Aug 10, 2025
04:25
Author Spotlight: Bridging Gaps in Anatomy and Establishing a Foundation for Algorithmic Studies
Published on: December 15, 2023
2.6K
Open-source, fully-automated hybrid cardiac substructure segmentation: development and optimisation.
Robert N Finnegan1,2,3, Vicky Chin4,5,6, Phillip Chlap4,5,6
1Northern Sydney Cancer Centre, Royal North Shore Hospital, St Leonards, NSW, Australia. robert.finnegan@sydney.edu.au.
Physical and Engineering Sciences in Medicine
|February 13, 2023
Summary
A new deep learning model accurately segments 18 cardiac substructures on CT scans for radiotherapy planning. This automated method improves cardiac dose evaluation and risk assessment in cancer patients.
Area of Science:
- Medical Imaging
- Radiotherapy Physics
- Computational Anatomy
Background:
- Radiotherapy for thoracic and breast cancers can cause cardiotoxicity.
- Cardiac substructure doses are increasingly recognized as predictive of specific outcomes.
- Accurate segmentation of cardiac substructures is crucial for developing clinical planning constraints but is currently lacking.
Purpose of the Study:
- To develop a novel, automated model for accurate and anatomically consistent segmentation of 18 cardiac substructures on CT scans.
- To provide quantitative data for developing clinical planning constraints in radiotherapy.
- To facilitate precise evaluation of cardiac doses and associated risks.
Main Methods:
- A multi-stage deep learning (DL) model combined with multi-atlas mapping and geometric modeling was developed.
- The model automatically segmented the whole heart, chambers, great vessels, valves, coronary arteries, and conduction nodes.
- Segmentation performance was evaluated using Dice Similarity Coefficient (DSC), Mean Distance to Agreement (MDA), Hausdorff Distance (HD), and volume ratio on 30 manually contoured CT scans.
Main Results:
- The automated model demonstrated reliable and accurate segmentation across various cardiac substructures, including challenging cases.
- Median DSC ranged from 0.81-0.93 for heart and chambers, 0.43-0.76 for great vessels and nodes, and 0.22-0.53 for valves.
- Median MDA was below 6 mm, median HD ranged from 7.7-19.7 mm, and median volume ratio was close to one for most structures.
- The fully automatic algorithm processed each case in 9-23 minutes.
Conclusions:
- The proposed fully-automatic method accurately delineates cardiac substructures on radiotherapy planning CT scans.
- Robust and anatomically consistent segmentations, especially for smaller structures, are a significant advantage.
- The open-source software will enable more precise cardiac dose and risk evaluations from clinical datasets.
Keywords:
Breast cancerCardiac substructuresCardiotoxicityDeep learningImage segmentationLung cancerRadiotherapy
