Echo-Mamba: A Lightweight Mamba-Based Framework for Generalizable Left Ventricular Segmentation Across Multiple...

Area of Science:

• Medical imaging
• Cardiovascular disease diagnosis
• Artificial intelligence in healthcare

Background:

• Accurate left ventricle (LV) segmentation in echocardiography is vital for assessing cardiac function and diagnosing cardiovascular disease.
• Existing segmentation models using Convolutional Neural Networks (CNNs) and Transformers have limitations in long-range modeling and computational complexity, hindering deployment on resource-constrained devices.
• There is a clinical need for lightweight, accurate segmentation solutions for echocardiography.

Purpose of the Study:

• To introduce Echo-Mamba, a novel lightweight architecture for left ventricular echocardiography segmentation.
• To address the limitations of existing models by enabling long-range dependency modeling with linear computational complexity.
• To improve adaptability to patient variations using an Adaptive Feature Enhancement (AFE) module.

Main Methods:

• Developed Echo-Mamba, a lightweight architecture utilizing a Mamba-based approach for efficient long-range dependency modeling.
• Incorporated an Adaptive Feature Enhancement (AFE) module to dynamically adjust for inter-patient variations in cardiac morphology.
• Evaluated Echo-Mamba against state-of-the-art methods on four public echocardiography datasets (EchoNet-Dynamic, EchoNet-Pediatric, CAMUS, HMC-QU).

Main Results:

• Echo-Mamba achieved generalizable segmentation performance across diverse anatomical targets, age groups, and multiview echocardiography.
• The model maintained a lightweight architecture with only 0.323M parameters, demonstrating efficiency.
• Demonstrated strong accuracy comparable to or exceeding existing state-of-the-art methods.

Conclusions:

• Echo-Mamba provides an effective and efficient solution for left ventricle segmentation in echocardiography.
• The proposed architecture meets the clinical demand for lightweight, accurate segmentation on low-cost medical devices.
• The study highlights the potential of Mamba-based architectures for medical image analysis, particularly in resource-constrained settings.