Machine-learning-based prediction of respiratory flow and lung volume from real-time cardiac MRI using MR-compatible...

Area of Science:

• Cardiovascular Imaging
• Respiratory Physiology
• Machine Learning in Medicine

Background:

• Cardiac real-time MRI (RT-MRI) combined with MR-compatible spirometry (MRcS) enables detailed study of heart-lung interactions.
• MRcS provides quantitative respiratory data, but shorter scan times are desirable, especially for vulnerable patients.

Purpose of the Study:

• To predict airflow and lung volume from RT-MR images after a short learning phase using combined RT-MRI and MRcS.
• To generate respiratory data for subsequent cardiac volumetries.

Main Methods:

• Adult healthy subjects (n=10) underwent cardiac RT-MRI with MRcS during free breathing.
• Optical flow estimation using the Lucas-Kanade method was applied to MR images.
• A ridge regression model predicted airflow and lung volume from optical flow, validated with cross-validation and a test dataset.

Main Results:

• A machine learning algorithm trained with MRcS data accurately predicted respiratory volume and flow.
• An optimized protocol achieved a balance between predictive power and training time (100s total duration).
• Predictions showed high accuracy (R² > 0.98 for airflow, R² > 0.99 for volume) with minimal error.

Conclusions:

• Machine learning enables feasible prediction of respiratory flow and lung volume from cardiac RT-MR images.
• This approach shortens MRcS time while ensuring accurate respiratory data acquisition during RT-MRI.