A Rapid, Accurate and Machine-Agnostic Segmentation and Quantification Method for CT-Based COVID-19 Diagnosis

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computational Biology

Background:

• COVID-19 pandemic necessitates advanced diagnostic tools beyond nucleic acid tests.
• Computed tomography (CT) imaging is crucial for COVID-19 diagnosis due to high false negative rates in early nucleic acid detection.
• Existing AI methods for CT-based COVID-19 diagnosis often require significant human intervention for segmentation.

Purpose of the Study:

• To develop a fully-automatic, rapid, and accurate AI method for segmenting and quantifying COVID-19 infection regions in CT scans.
• To overcome data scarcity issues in COVID-19 CT imaging through a novel simulation approach.
• To improve segmentation accuracy and reduce model complexity for large-scene-small-object problems in 3D CT data.

Main Methods:

• Development of the first CT scan simulator for COVID-19, modeling dynamic changes from real patient data.
• Implementation of a novel deep learning algorithm that decomposes 3D segmentation into three 2D problems.
• Validation of the method on multi-country, multi-hospital, and multi-machine datasets.

Main Results:

• The proposed method achieves superior performance in segmenting and quantifying COVID-19 infection regions compared to existing techniques.
• The CT scan simulator effectively addresses the challenge of limited COVID-19 patient data.
• The deep learning algorithm significantly improves segmentation accuracy while reducing model complexity.

Conclusions:

• The developed AI method offers a fully-automatic, rapid, and accurate solution for COVID-19 CT image analysis.
• This approach demonstrates significant application value in combating the COVID-19 pandemic by enhancing diagnostic capabilities.
• The method's machine-agnostic nature ensures broad applicability across different CT imaging sources.