Automatic cardiac cine MRI segmentation and heart disease classification.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Cardiology

Background:

• Cardiac cine MRI is crucial for non-invasive cardiovascular assessment.
• Fully convolutional neural networks (CNNs) enable pixel-wise classification for segmentation.
• Accurate segmentation of cardiac structures is vital for disease prediction.

Purpose of the Study:

• To develop an automated pipeline for cardiac segmentation and diagnosis using deep learning.
• To segment left ventricle cavity, right ventricle cavity, and left ventricle myocardium.
• To classify heart diseases based on volumetric features.

Main Methods:

• Utilized a UNet CNN variant for segmenting cardiac structures in short-axis cine MRI sequences.
• Implemented a signal processing approach for heart region localization and ROI cropping.
• Employed a classifier ensemble (MLP, Random Forest, SVM) for disease classification.

Main Results:

• Achieved a mean Dice overlap coefficient of 0.92 for segmenting three cardiac structures.
• Demonstrated good limits of agreement for derived clinical indices.
• Obtained an accuracy of 0.92 for heart disease classification on unseen data.

Conclusions:

• The proposed automated pipeline demonstrates near state-of-the-art performance in cardiac segmentation and disease classification.
• Deep learning and ensemble methods offer a robust approach for automated cardiac analysis.
• This pipeline has the potential to improve non-invasive cardiovascular disease assessment.