Efficient COVID-19 Segmentation from CT Slices Exploiting Semantic Segmentation with Integrated Attention Mechanism.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Coronavirus disease (COVID-19) is a global pandemic causing severe pneumonia.
• Computed tomography (CT) is crucial for COVID-19 screening, revealing characteristic lung patterns.
• Accurate segmentation of COVID-19 in CT images is vital for diagnosis and monitoring.

Purpose of the Study:

• To develop an automated method for segmenting COVID-19 regions in CT images.
• To enhance segmentation accuracy using an attention gate (AG) mechanism within a SegNet architecture.
• To evaluate the network's performance using various loss functions and cross-validation.

Main Methods:

• A SegNet-based deep learning model incorporating an attention gate (AG) mechanism was proposed.
• The network was trained and validated on a database of 473 COVID-19 CT images.
• Performance was assessed using dice, Tversky, and focal Tversky loss functions with fivefold cross-validation.

Main Results:

• The proposed model achieved high performance metrics: 92.73% sensitivity, 99.51% specificity, and 89.61% dice score.
• Attention gates improved model precision and predictive accuracy with minimal computational overhead.
• The method demonstrated superiority compared to existing studies in COVID-19 CT segmentation.

Conclusions:

• The developed attention-gated SegNet provides an accurate and efficient tool for automatic COVID-19 segmentation in CT images.
• This automated approach can aid clinicians in rapid diagnosis and disease staging.
• The model shows potential as an auxiliary tool in managing the COVID-19 pandemic.