MRL-Net: Multi-Scale Representation Learning Network for COVID-19 Lung CT Image Segmentation.

Area of Science:

• Medical Imaging
• Computer Vision
• Artificial Intelligence

Background:

• Accurate segmentation of COVID-19 lesions in lung CT images is crucial for patient screening and diagnosis.
• Challenges in segmentation include blurred, inconsistent lesion shapes and locations.

Purpose of the Study:

• To propose a novel Multi-Scale Representation Learning Network (MRL-Net) for improved COVID-19 lesion segmentation.
• To enhance feature representation by integrating Convolutional Neural Networks (CNN) and Transformer architectures.

Main Methods:

• MRL-Net integrates CNN and Transformer using Dual Multi-interaction Attention (DMA) and Dual Boundary Attention (DBA).
• DMA fuses multi-scale local detailed features (CNN) with global contextual information (Transformer).
• DBA focuses on lesion boundary features to improve representational learning.

Main Results:

• MRL-Net demonstrates superior performance compared to current state-of-the-art methods.
• The network achieves enhanced COVID-19 image segmentation accuracy.

Conclusions:

• MRL-Net effectively addresses the challenges of segmenting COVID-19 lesions in lung CT images.
• The proposed attention mechanisms significantly improve segmentation performance, aiding in clinical diagnosis.