MLCA2F: Multi-Level Context Attentional Feature Fusion for COVID-19 lesion segmentation from CT scans

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Accurate segmentation of Coronavirus disease 2019 (COVID-19) lesions in Computed Tomography (CT) images is crucial for diagnosis and treatment planning.
• Significant variations in lesion size, shape, position, boundary ambiguity, and complex structures pose challenges for existing segmentation methods.

Purpose of the Study:

• To develop a robust deep learning model for precise COVID-19 lesion segmentation in CT images.
• To introduce a novel multi-level contextual information fusion strategy to overcome segmentation challenges.

Main Methods:

• A deep learning model, Multi-Level Context Attentional Feature Fusion (MLCA2F), utilizing Multi-Scale Context-Attention Network (MSCA-Net) blocks.
• Integration of Multi-Scale Contextual Feature Fusion (MC2F) and Multi-Context Attentional Feature (MCAF) within MSCA-Net blocks.
• Extensive experiments on the Kaggle CT dataset to optimize the MLCA2F architecture.

Main Results:

• The proposed MLCA2F model demonstrated efficient and accurate segmentation of COVID-19 lesions.
• Comparative experiments showed superior performance against current state-of-the-art segmentation methods.
• The model effectively learned lesion details and improved boundary estimation of infected regions.

Conclusions:

• The MLCA2F framework offers a significant advancement in automated segmentation of COVID-19 lesions from CT scans.
• This deep learning approach holds substantial potential for enhancing clinical decision-making in COVID-19 management.
• The developed model addresses key challenges in lesion segmentation, paving the way for improved diagnostic tools.