Convolutional neural network regression for short-axis left ventricle segmentation in cardiac cine MR sequences

Area of Science:

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

Background:

• Automated left ventricular (LV) segmentation is vital for assessing cardiac function and morphology.
• Accurate segmentation aids in managing cardiac pathologies.
• Current methods face challenges in efficiency and accuracy.

Purpose of the Study:

• To develop and validate a novel automated LV segmentation method using convolutional neural network (CNN) regression.
• To parameterize LV segmentation in polar space using radial distances.
• To leverage domain-specific physical constraints within the CNN regression framework.

Main Methods:

• Parameterizing LV segmentation by radial distances in polar coordinates.
• Employing CNN regression to infer these parameters from cardiac MRI.
• Benchmarking against the Left Ventricle Segmentation Challenge (LVSC) dataset.
• Evaluating general applicability on the Kaggle Second Annual Data Science Bowl dataset.

Main Results:

• Achieved a Jaccard index of 0.77 on the LVSC dataset, outperforming existing automated methods.
• Obtained a Continuous Ranked Probability Score (CRPS) of 0.0124 on the Kaggle dataset, indicating strong performance in predicting LV volume.
• Demonstrated superior performance compared to other automated segmentation algorithms.

Conclusions:

• CNN regression combined with domain-specific features offers an effective approach for clinical cardiac segmentation.
• The proposed method provides accurate and efficient automated LV segmentation.
• This technique has the potential to significantly improve cardiac function quantification and patient management.