Toward Right Ventricle Segmentation in Cardiac MRIs via Feature Multiplexing and Multiscale Weighted Convolution

Area of Science:

• Medical Imaging
• Cardiology
• Artificial Intelligence

Background:

• Cardiovascular diseases are a leading cause of mortality worldwide.
• Accurate segmentation of cardiac structures, particularly the right ventricle (RV), in cardiac magnetic resonance images (MRIs) is essential for diagnosis and treatment.
• Current automated RV segmentation methods face challenges due to the RV's complex anatomy and image characteristics.

Purpose of the Study:

• To propose a novel triple-path segmentation model, FMMsWC, for accurate automated right ventricle segmentation in cardiac MRIs.
• To introduce and evaluate two new modules: feature multiplexing (FM) and multiscale weighted convolution (MsWC).
• To demonstrate the model's effectiveness on benchmark datasets and compare it with existing state-of-the-art methods.

Main Methods:

• Development of the FMMsWC model, a triple-path deep learning architecture.
• Integration of feature multiplexing (FM) and multiscale weighted convolution (MsWC) modules for enhanced feature extraction.
• Validation on two public cardiac MRI datasets: MICCAI2017 ACDC and M&MS.
• Comparative analysis against leading automated segmentation techniques.

Main Results:

• The FMMsWC model achieved superior performance in RV segmentation compared to state-of-the-art methods.
• Segmentation accuracy demonstrated by FMMsWC closely approximated that of manual segmentations by clinical experts.
• The model facilitates precise cardiac index measurement for rapid cardiac function assessment.

Conclusions:

• The proposed FMMsWC model offers a significant advancement in automated RV segmentation from cardiac MRIs.
• This improved segmentation accuracy aids in the diagnosis and treatment of cardiovascular diseases.
• The FMMsWC model shows substantial potential for clinical application in cardiology.