Cardiac disease discrimination from 3D-convolutional kinematic patterns on cine-MRI sequences

Area of Science:

• Cardiology
• Medical Imaging
• Artificial Intelligence

Background:

• Cine-magnetic resonance imaging (cine-MRI) is crucial for visualizing cardiac anatomy and detecting pathologies.
• Current cine-MRI analysis is subjective and prone to diagnostic errors.

Purpose of the Study:

• To develop a spatiotemporal model for classifying cardiac conditions based on kinematic movements.
• To improve the accuracy of cardiac disease diagnosis using advanced imaging analysis.

Main Methods:

• A 3D convolutional neural network was employed to analyze cardiac kinematic patterns.
• Kinematic maps were derived from apparent velocity maps computed using dense optical flow.
• The model learned to differentiate between various cardiac pathologies and normal cases.

Main Results:

• The multi-class classification achieved an average accuracy of 78.00% and a F1 score of 75.55%.
• Binary classification between pathologies and control cases reached 92.31% accuracy.
• The model successfully discriminated between myocardial infarction, dilated cardiomyopathy, hypertrophic cardiomyopathy, and abnormal right ventricle.

Conclusions:

• The proposed method effectively identifies abnormal kinematic patterns associated with cardiac pathologies.
• The learned spatiotemporal descriptors show promise as digital biomarkers for cardiac diseases.
• This approach can aid in the objective diagnosis and characterization of heart conditions.