Hypergraph learning for identification of COVID-19 with CT imaging.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computational Biology

Background:

• COVID-19 presents a significant global health challenge, necessitating rapid diagnostic tools.
• CT imaging aids in early COVID-19 screening, but differentiating it from community-acquired pneumonia (CAP) is difficult due to similar presentations.
• Accurate differentiation between COVID-19 and CAP is crucial for effective patient management and public health strategies.

Purpose of the Study:

• To develop and evaluate a novel method for accurately identifying COVID-19 cases from community-acquired pneumonia (CAP) using CT imaging.
• To address the challenge of distinguishing between COVID-19 and CAP cases with similar clinical and imaging features.
• To propose an Uncertainty Vertex-weighted Hypergraph Learning (UVHL) method for improved diagnostic accuracy.

Main Methods:

• Extraction of diverse features, including regional and radiomics features, from CT images of COVID-19 and CAP cases.
• Formulation of case relationships using a hypergraph structure, where each case is a vertex.
• Integration of vertex uncertainty scores as weights within the hypergraph for a robust learning process.
• Application of a vertex-weighted hypergraph learning model for predicting COVID-19 in new cases.

Main Results:

• The proposed UVHL method demonstrated high effectiveness and robustness in identifying COVID-19 from CAP using CT images.
• Experimental validation on a large, multi-center dataset comprising 2148 COVID-19 and 1182 CAP cases.
• The UVHL method outperformed existing state-of-the-art methods in COVID-19 identification accuracy.
• The approach successfully models the complex relationships and uncertainties inherent in differentiating COVID-19 from CAP.

Conclusions:

• The Uncertainty Vertex-weighted Hypergraph Learning (UVHL) method offers a powerful and accurate approach for differentiating COVID-19 from CAP based on CT imaging.
• The study highlights the potential of hypergraph learning and uncertainty modeling in medical image analysis for infectious disease diagnosis.
• The findings support the clinical utility of the UVHL method for early and reliable screening of COVID-19, especially in resource-limited settings.
• Further research could explore the integration of additional data modalities to enhance diagnostic performance.