Research Interest (V): Development and evaluation of AI-based COVID-19 pneumonia diagnosis

I developed and evaluated an AI-based COVID-19 pneumonia diagnosis model [1]. I developed an ensemble AI model using multi-center, multi-vendor chest x-ray images that outperformed three radiologists (AUC 0.94 vs 0.85).

Notable Citation

CV19-Net inspired scholars worldwide to implement, analyze, and use it as an example for future developments. Roberts et al. [2] performed a large-scale analysis of 2,215 related COVID-19 classification studies to assess their bias and clinical utility. CV19-Net was the only 2 out of 2,215 works that were considered without high risks of bias shown in their Table 1 and the only 4 out of 2215 with a large and balanced testing dataset.

Dr Summers, Chief of Clinical Image Processing Service and Chief of Imaging Biomarkers and Computer-Aided Diagnosis Laboratory from the National Institutes of Health Clinical Center wrote an editorial [3] in the Radiology journal after the spikes of AI-based COVID-19 related works. He used CV19-Net among thousands of other works on this topic as the illustrative example of binary COVID-19 diagnosis.

Clinical Impact

The wide recognition of CV19-Net led to its clinical translation into the UW Health clinical workflow and continues to provide decision support for radiologists to prioritize high-risk patients.

Reference

[1] Ran Zhang, Xin Tie, Zhihua Qi, Nicholas B Bevins, Chengzhu Zhang, and others. “Diagnosis of coronavirus disease 2019 pneumonia by using chest radiography: Value of artificial intelligence.” Radiology 298.2 (2021): E88-E97.

[2] Roberts M, Driggs D, Thorpe M, et al. “Common pitfalls and recommendations for using machine learning to detect and prognosticate for COVID-19 using chest radiographs and CT scans”. Nature Machine Intelligence, 2021, 3(3): 199-217.

[3] R M Summers. Artificial intelligence of COVID-19 imaging: a hammer in search of a nail [J]. Radiology, 2021.