Automatic COVID-19 CT segmentation using U-Net integrated spatial and channel attention mechanism.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• The COVID-19 pandemic significantly impacted global public health.
• Computed Tomography (CT) is crucial for COVID-19 screening and patient monitoring.
• Accurate and rapid segmentation of COVID-19 lesions in CT images is essential for clinical decision-making.

Purpose of the Study:

• To develop an efficient and accurate deep learning model for segmenting COVID-19 in CT scans.
• To enhance feature representation in U-Net architectures using attention mechanisms.
• To improve the segmentation of small lesions using a specialized loss function.

Main Methods:

• A U-Net based segmentation network incorporating spatial and channel attention modules was proposed.
• Attention mechanisms were used to re-weight feature representations for capturing contextual relationships.
• Focal Tversky loss was implemented to address the challenge of segmenting small lesions.

Main Results:

• The proposed method demonstrated accurate and rapid segmentation of COVID-19 in CT images.
• Segmentation of a single CT slice was achieved in just 0.29 seconds.
• The model attained a Dice Score of 83.1% and a Hausdorff Distance of 18.8.

Conclusions:

• The attention-based U-Net model provides an effective solution for COVID-19 segmentation in CT scans.
• The integration of attention mechanisms and focal Tversky loss enhances segmentation performance.
• The method shows promise for clinical applications in COVID-19 diagnosis and monitoring.