Classification of COVID-19 in chest radiographs: assessing the impact of imaging parameters using clinical and simulated images

Loading...
Thumbnail Image

Date

2021-02-15

Journal Title

Journal ISSN

Volume Title

Repository Usage Stats

83
views
46
downloads

Citation Stats

Abstract

As computer-aided diagnostics develop to address new challenges in medical imaging, including emerging diseases such as COVID-19, the initial development is hampered by availability of imaging data. Deep learning algorithms are particularly notorious for performance that tends to improve proportionally to the amount of available data. Simulated images, as available through advanced virtual trials, may present an alternative in data-constrained applications. We begin with our previously trained COVID-19 x-ray classification model (denoted as CVX) that leveraged additional training with existing pre-pandemic chest radiographs to improve classification performance in a set of COVID-19 chest radiographs. The CVX model achieves demonstrably better performance on clinical images compared to an equivalent model that applies standard transfer learning from ImageNet weights. The higher performing CVX model is then shown to generalize effectively to a set of simulated COVID-19 images, both quantitative comparisons of AUCs from clinical to simulated image sets, but also in a qualitative sense where saliency map patterns are consistent when compared between sets. We then stratify the classification results in simulated images to examine dependencies in imaging parameters when patient features are constant. Simulated images show promise in optimizing imaging parameters for accurate classification in data-constrained applications.

Department

Description

Provenance

Subjects

Citation

Published Version (Please cite this version)

10.1117/12.2582223

Publication Info

Fricks, Rafael, Ehsan Abadi, Francesco Ria and Ehsan Samei (2021). Classification of COVID-19 in chest radiographs: assessing the impact of imaging parameters using clinical and simulated images. Medical Imaging 2021: Computer-Aided Diagnosis, 115970A. pp. 1–11. 10.1117/12.2582223 Retrieved from https://hdl.handle.net/10161/22421.

This is constructed from limited available data and may be imprecise. To cite this article, please review & use the official citation provided by the journal.

Scholars@Duke

Abadi

Ehsan Abadi

Assistant Professor in Radiology

Ehsan Abadi, PhD is an imaging scientist at Duke University. He serves as an Assistant Professor in the departments of Radiology and Electrical & Computer Engineering, a faculty member in the Medical Physics Graduate Program and Carl E. Ravin Advanced Imaging Laboratories, and a co-Lead in the Center for Virtual Imaging Trials. Ehsan’s research focuses on quantitative imaging and optimization, CT imaging, lung diseases, computational human modeling, and medical imaging simulation. He is actively involved in developing computational anthropomorphic models with various diseases such as COPD, and scanner-specific simulation platforms (e.g., DukeSim) for imaging systems. Currently, his work is centered on identifying and optimizing imaging systems to ensure accurate and precise quantifications of lung diseases.

Ria

Francesco Ria

Assistant Professor of Radiology

Dr. Francesco Ria is a medical physicist and he serves as an Assistant Professor in the Department of Radiology. Francesco has an extensive expertise in the assessment of procedure performances in radiology. In particular, his research activities focus on the simultaneous evaluation of radiation dose and image quality in vivo in computed tomography providing a comprehensive evaluation of radiological exams. Moreover, Francesco is developing and investigating novel mathematical models that, uniquely in the radiology field, can incorporate a comprehensive and quantitative risk-to-benefit assessment of the procedures; he is continuing to apply his expertise towards the definition of new patient specific risk metrics, and in the assessment of image quality in vivo also using state-of-the-art imaging technology, such as photon counting computed tomography scanners, and machine learning reconstruction algorithms.

Dr. Ria is a member of the American Association of Physicists in Medicine task group 392 (Investigation and Quality Control of Automatic Exposure Control System in CT), of the American Association of Physicists in Medicine Public Education working group (WGATE), and of the Italian Association of Medical Physics task group Dose Monitoring in Diagnostic Imaging.

Samei

Ehsan Samei

Reed and Martha Rice Distinguished Professor of Radiology

Dr. Ehsan Samei, PhD, DABR, FAAPM, FSPIE, FAIMBE, FIOMP, FACR is a Persian-American medical physicist. He is the Reed and Martha Rice Distinguished Professor of Radiology, and Professor of Medical Physics, Biomedical Engineering, Physics, and Electrical and Computer Engineering at Duke University. He serves as the Chief Imaging Physicist for Duke University Health System, the Director of the Carl E Ravin Advanced Imaging Laboratories and the Center for Virtual Imaging Trials (CVIT), and co-PI of one the five Centers of Excellence in Regulatory Science and Innovation (CERSI), Triangle CERSI. He is certified by the American Board of Radiology, recognized as a Distinguished Investigator by the Academy of Radiology Research, and awarded Fellow by five professional organization. He founded/co-founded the Duke Medical Physics Program, the Duke Imaging Physics Residency Program, the Duke Clinical Imaging Physics Group, the Center for Virtual Imaging Trials, and the Society of Directors of Academic Medical Physics Programs (SDAMPP). He has held senior leadership positions in the AAPM, SPIE, SDAMPP, and RSNA, including election to the presidency of the SEAAPM (2010-2011), SDAMPP (2011), and AAPM (2023).

Dr. Samei’s scientific expertise include x-ray imaging, theoretical imaging models, simulation methods, and experimental techniques in medical image formation, quantification, and perception.  His research aims to bridge the gap between scientific scholarship and clinical practice, in the meaningful realization of translational research, and in clinical processes that are informed by scientific evidence. He has advanced image quality and safety metrics and radiometrics that are clinically relevant and that can be used to design, optimize, and monitor interpretive and quantitative performance of imaging techniques. These have been implemented in advanced imaging performance characterization, procedural optimization, and clinical dose and quality analytics. His most recent research interests have been virtual clinical trial across a broad spectrum of oncologic, pulmonary, cardiac, and vascular diseases, and developing methodological advances that provide smart fusions of principle-informed and AI-based, data-informed approaches to scientific inquiry.

Dr. Samei has mentored over 140 trainees (graduate and postgraduate). He has >1400 scientific publications including >360 referred journal articles, ~600 conference presentations, and 4 books. Citations to his work is reflected in an h-index of 74 and a Weighted Relative Citation Ratio of 613. His laboratory of over 20 researchers has been supported continuously over two decades by 44 extramural grants, culminating in a NIH Program Project grant in 2021 to establish the national Center for Virtual Imaging Trials (CVIT), joining a small number of prominent Biomedical Technology Research Centers across the nation. In 2023, he, along with 3 other PIs, was awarded to lead one of five national Centers of Excellence in Regulatory Science and Innovation (Triangle CERSI) by the FDA.


Unless otherwise indicated, scholarly articles published by Duke faculty members are made available here with a CC-BY-NC (Creative Commons Attribution Non-Commercial) license, as enabled by the Duke Open Access Policy. If you wish to use the materials in ways not already permitted under CC-BY-NC, please consult the copyright owner. Other materials are made available here through the author’s grant of a non-exclusive license to make their work openly accessible.