Image-based estimation of the left ventricular cavity volume using deep learning and Gaussian process with...

Area of Science:

• Medical Imaging
• Cardiology
• Machine Learning

Background:

• Accurate estimation of left ventricular (LV) cavity volume is crucial for cardiac function assessment.
• Manual volume extraction from cardiac magnetic resonance (CMR) imaging is time-consuming and prone to inter-observer variability.

Purpose of the Study:

• To develop and validate an automated, image-based method for estimating LV cavity volumes from CMR data.
• To improve the accuracy of LV volume estimation compared to existing methods.
• To demonstrate a clinical application of automated volume estimation.

Main Methods:

• Development of an image-based method utilizing deep learning and Gaussian processes.
• Training a stepwise regression model on CMR data from 339 patients and healthy volunteers.
• Validation of the model for estimating LV volumes at end-diastole and end-systole.

Main Results:

• Reduced the root mean square error (RMSE) of LV cavity volume estimation from approximately 13 ml to 8 ml.
• Achieved an automated estimation error of 8 ml, compared to a manual measurement RMSE of 4 ml.
• Successfully inferred passive myocardial material properties using automated volume estimates and a cardiac model.

Conclusions:

• The developed automated method provides a significant improvement in LV volume estimation accuracy from CMR data.
• The method offers a time-efficient and reliable alternative to manual measurements.
• Inferred myocardial material properties hold potential for patient-specific diagnosis and treatment planning.