Automatic left ventricle volume calculation with explainability through a deep learning weak-supervision methodology.

Area of Science:

• Cardiovascular Imaging
• Medical Image Analysis
• Artificial Intelligence in Medicine

Background:

• Left ventricle volume assessment is crucial for diagnosing cardiac pathologies.
• Accurate volume quantification from cardiac MRI is essential for clinical decision-making.

Purpose of the Study:

• To develop a neural network for direct estimation of left ventricle volume from short-axis cine MRI.
• To provide explainable segmentation masks for the estimated left ventricle volumes.
• To enable easier training of neural networks when segmentation labels are scarce.

Main Methods:

• A 3D U-net architecture with a scanning module was employed for weakly-supervised learning.
• The network was trained to estimate left ventricle volumes and object circularity.
• No labeled segmentation masks were required for training.

Main Results:

• The model achieved a mean relative error of 8% and a mean absolute error of 9.12 ml for volume estimation.
• A Pearson correlation coefficient of 0.95 was observed between real and estimated volumes.
• The generated segmentation masks had a mean Dice coefficient of 0.79.

Conclusions:

• The proposed method accurately estimates left ventricle end-diastolic volume using cardiac MRI.
• The network provides segmentation masks for explainability without needing segmentation labels.
• This approach offers a more trustworthy and easily trainable solution for clinicians and researchers.