Multi-center Integrating Radiomics, Structured Reports, and Machine Learning Algorithms for Assisted Classification...

Area of Science:

• Radiology
• Artificial Intelligence
• Medical Imaging

Background:

• Distinguishing Coronavirus Disease 2019 (COVID-19) pneumonia from other pneumonias on chest computed tomography (CT) scans is clinically important.
• Radiomics and machine learning (ML) offer potential tools for improving diagnostic accuracy.

Purpose of the Study:

• To evaluate the classification performance of structured report features, radiomics, and ML models in differentiating COVID-19 from other pneumonias using chest CT.
• To assess the impact of ML assistance on physician diagnostic performance.

Main Methods:

• A cohort of 64 COVID-19 and 64 non-COVID-19 pneumonia subjects was analyzed.
• Data was split into training (n=73) and validation (n=55) sets for model development and testing.
• Physicians performed CT readings with and without ML support; sensitivity, specificity, and inter-rater reliability (Cohen's Kappa) were calculated.

Main Results:

• Physician performance without ML: mean sensitivity 83.4%, specificity 64.3%.
• Physician performance with ML assistance: mean sensitivity 87.1%, specificity 91.1%.
• ML significantly improved inter-rater reliability from moderate to substantial.

Conclusions:

• Integrating structured reports and radiomics with ML shows promise for assisted COVID-19 classification on chest CT scans.
• ML-assisted interpretation can enhance diagnostic accuracy and consistency in identifying COVID-19 pneumonia.