Optical Flow-Guided Cine MRI Segmentation With Learned Corrections.

Area of Science:

• Medical Imaging
• Cardiovascular Imaging
• Artificial Intelligence in Medicine

Background:

• Cardiac cine MRI is crucial for diagnosing heart conditions, requiring analysis of temporal changes in heart structures.
• Accurate segmentation of regions of interest (ROIs) like ventricles and atria is essential for clinical assessment.
• Manual segmentation is time-consuming and prone to inter-observer variability.

Purpose of the Study:

• To develop an automated method for propagating segmented regions of interest (ROIs) throughout the cardiac cycle in cardiac cine MRI.
• To improve the accuracy and efficiency of cardiac structure segmentation for clinical diagnosis.
• To enable precise tracking of cardiac motion and volume changes.

Main Methods:

• A novel approach combining 3D TV-L1 optical flow for motion estimation with a 3D U-Net convolutional neural network (CNN) for segmentation refinement.
• Bidirectional optical flow is used to propagate expert-annotated masks from end-systole and end-diastole phases.
• A custom loss function, including a penalization term, is employed to refine segmentation masks and ensure consistency.

Main Results:

• The proposed method successfully automates the propagation of ROIs across the cardiac cycle.
• Benchmarking on dedicated and public datasets demonstrates state-of-the-art performance.
• The novel loss function facilitates single-patient fine-tuning, enhancing segmentation accuracy.

Conclusions:

• The developed automated segmentation method offers a significant advancement in cardiac MRI analysis.
• This technique provides accurate and efficient tracking of cardiac structures, aiding clinical diagnosis.
• The approach holds promise for improving quantitative assessment of cardiac function.