A Joint Classification Method for COVID-19 Lesions Based on Deep Learning and Radiomics

Area of Science:

• Medical imaging analysis
• Artificial intelligence in healthcare
• Infectious disease diagnostics

Background:

• Novel coronavirus pneumonia (COVID-19) is a rapidly spreading acute respiratory illness posing significant global public health challenges.
• Accurate classification of lung diseases is crucial for prognosis and timely clinical management.
• Deep learning and radiomics offer potential for enhanced diagnostic capabilities in medical imaging.

Purpose of the Study:

• To evaluate the performance of deep learning and radiomics in classifying COVID-19 lung lesions.
• To identify key image characteristics associated with COVID-19 lung disease.
• To compare the classification efficacy of deep features, radiomics features, and their combination.

Main Methods:

• Development of a Multi-Feature Pyramid Network (MFPN) for extracting deep lesion features.
• Application of six machine learning algorithms for classification tasks.
• Comparative analysis of classification performance using deep features, radiomics features, and combined features.

Main Results:

• The combined approach utilizing both radiomics and deep features demonstrated superior performance in classifying COVID-19 lung lesions.
• The MFPN model effectively extracted relevant deep features for lesion characterization.
• The study confirmed the clinical utility of integrating radiomics and deep learning for COVID-19 diagnosis.

Conclusions:

• Integrating radiomics and deep learning methods provides a robust approach for COVID-19 image classification.
• This combined strategy offers valuable clinical application potential for improved patient management.
• Further research can explore these AI-driven methods for other pulmonary diseases.