Related Experiment Video
Updated: Apr 28, 2026
11:38
Voluntary Breath-hold Technique for Reducing Heart Dose in Left Breast Radiotherapy
Published on: July 3, 2014
47.8K
Machine Learning-based Prediction of Mean Heart Dose and Deep Inspiration Breath-hold Selection in Left-sided Breast
Deepali Patil1, Mukesh Kumar Zope1, Rishi Raj2
1Department of Medical Physics and State Cancer Institute, Indira Gandhi Institute of Medical Sciences, Patna, Bihar, India.
Journal of Medical Physics
|April 27, 2026
Summary
Machine learning accurately predicts heart dose in left-sided breast cancer patients, identifying those who benefit from deep inspiration breath-hold (DIBH) for cardiac sparing. This tool uses simple anatomical metrics from free-breathing scans to guide DIBH selection.
Area of Science:
- Radiation Oncology
- Medical Physics
- Machine Learning in Healthcare
Background:
- Deep inspiration breath-hold (DIBH) is effective in reducing cardiac radiation exposure for left-sided breast cancer patients.
- Patient selection is crucial for DIBH implementation due to resource limitations.
- Accurate prediction of heart mean dose is needed to identify suitable candidates for DIBH.
Purpose of the Study:
- To develop and validate a machine learning tool for predicting heart mean dose in left-sided breast cancer patients.
- To identify patients who would benefit from DIBH using simple anatomical predictors from free-breathing (FB) scans.
- To assess the tool's performance across different volumetric modulated arc therapy (VMAT) techniques.
Main Methods:
- Retrospective analysis of 120 left-sided breast cancer patients' VMAT treatment plans (VMAT-2P, VMAT-4P, VMAT-5P) on FB scans.
- Measurement of anatomical predictors: maximum heart distance (MHD) and heart-to-PTV distance (HPD).
- Application of Elastic Net regression for continuous dose prediction and logistic regression for binary classification of DIBH necessity (threshold: 5 Gy heart mean dose).
- Validation using an independent cohort (n=25) with paired FB-DIBH scans and a test cohort (n=24) from the development dataset.
Main Results:
- DIBH reduced mean heart dose by 34% and decreased high-risk patients by 69%-80% in the validation cohort.
- Strong correlations were found between FB predictions and DIBH doses for VMAT-2P (r=0.667-0.720).
- The machine learning model achieved mean absolute errors of 0.81-1.02 Gy and demonstrated high accuracy (87.5%), sensitivity (83%-92%), and specificity (83%-92%) for VMAT-2P and VMAT-4P.
- VMAT-5P showed reduced classification performance (58.3% accuracy).
Conclusions:
- Machine learning software accurately predicts mean heart dose during pre-treatment planning for left-sided breast cancer.
- The tool enables informed DIBH selection for cardiac sparing based on simple anatomical metrics from FB CT scans.
- This approach facilitates efficient patient selection for DIBH, optimizing radiation therapy planning.
