Focus, Fusion, and Rectify: Context-Aware Learning for COVID-19 Lung Infection Segmentation

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Pulmonology

Background:

• The COVID-19 pandemic necessitates rapid diagnostic tools due to limited resources.
• Computed tomography (CT) analysis shows comparable accuracy to PCR for COVID-19 detection.
• Automatic segmentation of lung infections in CT scans is crucial for efficient COVID-19 management.

Purpose of the Study:

• To develop an automated method for segmenting lung infections in CT scans.
• To address the challenges of high variability and indistinction in COVID-19 CT image appearances.
• To improve the speed and accuracy of COVID-19 diagnosis and follow-up using AI.

Main Methods:

• A novel context-aware neural network was proposed for lung infection segmentation.
• Autofocus and panorama modules were designed to extract detailed features and contextual information.
• A structure consistency rectification method was introduced for improved segmentation accuracy.

Main Results:

• The proposed method demonstrated effectiveness on both multiclass and single-class COVID-19 CT datasets.
• The mean intersection over union (mIoU) scores achieved were 64.8%, 65.2%, and 73.8% on benchmark datasets.
• The approach successfully segmented lung infections, highlighting its potential in clinical settings.

Conclusions:

• The developed context-aware neural network offers a robust solution for automated lung infection segmentation in CT scans.
• This AI-driven approach can significantly support rapid diagnosis, treatment, and monitoring of COVID-19 patients.
• The method's performance indicates its viability as a valuable tool in managing infectious respiratory diseases.