Novel Quantitative Tool for Assessing Pulmonary Disease Burden in COVID-19 Using Ultrasound
H Sagreiya1*, M Jacobs2, A Akhbardeh2, (1) University of Pennsylvania, Philadelphia, PA, (2) Johns Hopkins University, Baltimore, MD
Presentations
PO-GePV-M-62 (Sunday, 7/10/2022) [Eastern Time (GMT-4)]
ePoster Forums
Purpose: The outbreak of coronavirus (CV-19) occurred in late 2019, the WHO later declared it a worldwide pandemic, and it has led to over 6 million deaths. Radiological imaging via Point of Care Ultrasound (POCUS) and other modalities is playing a central role in its management. We analyzed serial POCUS in patients with COVID- 19 using computer vision and artificial intelligence methods. We developed and tested the current POCUS computer vision and machine learning technique, termed the COVID Lung Index (CLU), using 52 cases from different institutions and developed a novel quantitative biomarker termed the CLU index, a measure of the patient’s quantitative pulmonary disease burden.
Methods: The CLU technique employs an unsupervised, patient-specific model that does not require training on a large dataset. The patient-specific approach contrasts with conventional machine-learning/deep-learning methods, which require training. COVID Lung Index assesses various lung ultrasound features (A-lines, B-lines, consolidation, pleural effusion), resulting in a quantitative biomarker termed the CLU index, a measure of the patient’s quantitative disease burden within the lungs. Initial performance was assessed on publicly available lung ultrasound data in a sample size of 52 patients The POCUS data were scanned with multiple US devices, including Siemens, Philips, GE, Butterfly iQ, and SonoSite.
Results: Machine learning with CLU was 100% concordant with radiologist results for the following findings: A-lines (12), patchy B-lines (19), confluent and diffuse B-lines (17), thickened/irregular pleural lines (13), pleural effusion (6), subpleural consolidations (12), and consolidations with air bronchogram (9). This tool quantified pulmonary disease burden in these patients, with the CLU index varying with the extent of pulmonary disease as patients worsened and improved.
Conclusion: This technique identified pulmonary findings associated with CV-19 with high accuracy. It can quantify disease burden, being particularly helpful for disease extent, treatment effectiveness, and monitoring COVID-19 “long haulers.”
Funding Support, Disclosures, and Conflict of Interest: Hersh Sagreiya was funded by the Radiological Society of North America, grant number RSCH2028.
Keywords
Ultrasonics, Quantitative Imaging, Image Analysis
Taxonomy
IM- Ultrasound : Quantitative imaging
Contact Email