Computational Platform Based on Deep Learning for Segmenting Ventricular Endocardium in Long-axis Cardiac MR Imaging.

Area of Science:

• Cardiology
• Medical Imaging
• Artificial Intelligence

Background:

• Accurate segmentation of cardiac chambers in cardiovascular magnetic resonance (CMR) is crucial for diagnosing and monitoring heart conditions.
• Manual segmentation is time-consuming and prone to inter-observer variability.
• Deep learning (DL) offers potential for automated and efficient image analysis.

Purpose of the Study:

• To develop and validate an automated computational platform for left ventricular (LV) and right ventricular (RV) endocardium segmentation in long-axis cine CMR images.
• To assess the accuracy and speed of the proposed DL-based segmentation method.
• To evaluate the correlation of DL-derived cardiac function parameters with ground truth.

Main Methods:

• Utilized modified deep U-Net convolutional neural networks for endocardium segmentation.
• Trained the model on 4800 images from 40 human subjects (healthy and cardiac patients).
• Validated the model on 6000 images from 50 subjects.

Main Results:

• Achieved high segmentation accuracy with an average Dice metric of 0.929 ± 0.036 and Jaccard index of 0.869 ± 0.059.
• Demonstrated strong correlation with ground truth for LV ejection fraction (R=0.975) and fractional area change (LV: R=0.959–0.971, RV: R=0.927).
• Completed segmentation in less than 3 seconds per subject (< 30 ms/image).

Conclusions:

• The proposed DL framework provides a fully automated and rapid solution for LV and RV endocardium segmentation in long-axis cine CMR.
• The method shows high accuracy and strong agreement with manual segmentation and clinical parameters.
• This automated approach holds promise for efficient and reliable cardiac image analysis in clinical settings.