MSD-Net: Multi-Scale Discriminative Network for COVID-19 Lung Infection Segmentation on CT

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Pulmonology

Background:

• The COVID-19 pandemic necessitates rapid and accurate diagnostic tools.
• Computed tomography (CT) is a crucial imaging modality for COVID-19 diagnosis.
• Accurate segmentation of lung infections aids in patient management and treatment.

Purpose of the Study:

• To develop and evaluate a deep learning model for multi-class segmentation of COVID-19 lung infections on CT images.
• To improve the accuracy and efficiency of identifying ground-glass opacities, interstitial infiltrates, and consolidation related to COVID-19.

Main Methods:

• A multi-scale discriminative network (MSD-Net) was proposed, incorporating pyramid convolution blocks (PCB), channel attention blocks (CAB), and residual refinement blocks (RRB).
• The PCB enhances feature extraction for various infection sizes.
• CAB and RRB refine feature maps for improved segmentation accuracy.

Main Results:

• The MSD-Net achieved Dice Similarity Coefficients (DSC) of 0.7422, 0.7384, and 0.8769 for the three infection categories.
• Sensitivity and specificity results were promising, with values such as (0.8593, 0.9742) for one category.
• The network demonstrated effectiveness in segmenting diverse COVID-19 lung infection patterns.

Conclusions:

• The proposed MSD-Net effectively segments COVID-19 lung infections on CT images.
• This deep learning approach shows potential for assisting in the clinical diagnosis and treatment of COVID-19.
• The network's ability to segment different infection types aids in precise patient assessment.