Deep learning-based technique for lesions segmentation in CT scan images for COVID-19 prediction

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• COVID-19 has caused a global health crisis since 2019, overwhelming healthcare systems worldwide.
• Accurate diagnosis of COVID-19 is crucial for effective patient management and pandemic control.
• Chest CT imaging is vital for identifying COVID-19 related lung abnormalities like ground glass opacity and consolidation.

Purpose of the Study:

• To develop a novel deep learning application for automated COVID-19 segmentation and analysis in CT images.
• To enhance the diagnostic capabilities for COVID-19 by accurately identifying key pathological regions.
• To provide a robust tool for distinguishing COVID-19 related lung lesions from other pneumonias.

Main Methods:

• A deep learning system based on a context aggregation neural network was developed.
• The network incorporates three modules: context fuse model (CFM), attention mix module (AMM), and residual convolutional module (RCM).
• Experiments were conducted using the COVID-x-CT dataset for training and validation.

Main Results:

• The developed system accurately detected ground glass opacity and consolidation areas in CT images.
• The system demonstrated superior performance compared to existing state-of-the-art methods.
• Achieved an accuracy exceeding 96.23% in COVID-19 segmentation and analysis.

Conclusions:

• The proposed deep learning system offers a highly effective and accurate method for COVID-19 analysis in CT scans.
• The context aggregation neural network architecture proved successful in identifying critical lung lesions.
• This work contributes a valuable tool for improving COVID-19 diagnosis and patient care.