|
As we enter our third year of our contract with NIBIB, we wanted to highlight some notable MIDRC accomplishments
from the past two years:
- Ingesting over 150,000 imaging studies, with approximately half now undergoing curation and half already published on the data.midrc.org site, ready for AI investigators
- Creating the MIDRC Sequestered Data Commons, to perform task-based sampling and algorithm performance assessment, aid in regulatory approval, and to accelerate the translation of machine learning and AI to real-world uses in clinical
- Continuously tracking the diversity of MIDRC imaging data to better provide diverse, representative data to AI developers
- Facilitating cohort selection for MIDRC data users by implementing a mapping to match highly varied incoming DICOM Study Descriptions to Logical Observation Identifiers, Names and Codes (LOINC) values within the data portal
- Developing natural language processing (NLP) computer methods to extract information from radiology reports of associated medical images
- Creating, deidentifying, and normalizing data methods and tools being used by ACR and RSNA in their imaging and data collection
- Developing an imaging-based Data Dictionary with data elements along with SOPs for the data model and to enable cohort building on the MIDRC Gen3 user portal
- Developing methods of interoperability between MIDRC and non-imaging commons, such as NCATS N3C and NHLBI Biodata Catalyst PETAL
- Conducting MIDRC's first Grand Challenge, the COVIDx Challenge, a COVID classification Challenge (see more information below)
- Establishing a collaboration with Argonne National Laboratory on the development and training of privacy-preserving federated learning models using MIDRC data
- Investigating and informing MIDRC CT image quality by performing a multi-institutional imaging phantom study
- Building the MIDRC Metrics and Resources Portal, including MIDRC GitHub, assisting MIDRC data users with analysis techniques, references, software options and potential metrics for use in the evaluation of algorithm performance
- Analyzing mitigation strategies for various forms of data bias possible in model development, evaluation and deployment, and data collection, preparation, and annotation
- Developing AI algorithms for use in the diagnosis and prognosis of COVID-19 on chest radiographs and CT scans.
- Authoring articles on data science, artificial intelligence, and the imaging of COVID 19, read more
|
|
|
MIDRC continually publishes new data.
Registered users can run queries on and build cohorts of both chest radiographs and CT scans.
Tell us about your experience using the MIDRC Data Portal
|
|
Nov. 27 to Dec. 1, 2022 McCormick Place
Technical Exhibits: Nov. 27–30
|
|
|
Come visit our MIDRC exhibit space for demos on our data portal, annotation, MIDRC resources and tools, and to meet our investigators and team!
You can find us in the Lakeside Learning Center Informatics space, between the deep learning center and digital posters.
|
|
|
MIDRC at AAPM Annual Meeting, July 2022
|
|
We launched our first public
Grand Challenge this Fall,
the MIDRC COVIDx Challenge!
The goal of this Challenge was to train an AI/machine learning model
in the task of distinguishing between COVID-negative and COVID-positive patients using frontal-view portable chest radiographs (CXRs) from contributed MIDRC data, and both US and international teams registered to compete. Participants were encouraged to use publicly available data from the MIDRC Data Commons during the training phase of the Challenge, and instructions for building a training cohort were posted in the MIDRC GitHub repository. Data used for the testing and validation phases was not publicly available, however, and the competing teams’ final containerized (Docker) submissions were accepted on the MedICI Challenge
platform through the end of October.
Cash awards, generously sponsored by the International Society for Optics and Phototonics (SPIE), will be given to the top-ranked teams, who will also be announced during MIDRC’s presentation in the AI Innovation Theatre during the upcoming RSNA meeting (Chicago, IL, Nov 27 – Dec 1). The MIDRC Grand Challenge Working Group (GCWG) plans to announce its second Challenge in Spring 2023.
|
|
|
MIDRC Researcher Spotlights
|
|
|
|
Michael McNitt-Gray, PhD,
DABR, FAAPM, FACR
Mike received his BSEE from Washington University in St. Louis, and
his MSEE from Carnegie-Mellon University. After working in the electric power and alternative energy industry for seven years, he returned to graduate school and completed his PhD in Biomedical Physics from UCLA. He remained there and is now a Professor in the Department of Radiological Sciences at UCLA and does research in the areas of computed tomography image acquisition, physics and image analysis, primarily using image processing and machine learning techniques to analyze radiological images. He has been the Director of the UCLA Physics and Biology in Medicine (PBM) graduate program for more than 18 years and continues to teach in that program and supervise graduate students. He is certified by the American Board of Radiology in the field of Diagnostic Medical Physics and is a Fellow of the American Association of Physicists in Medicine (AAPM) as well as a Fellow of the American College of Radiology (ACR).
As the lead investigator of MIDRC's technology development project (TDP)
3c research group, he's coordinating the development of an online tool for MIDRC data users, the MIDRC Metrics and Resources portal, which helps guide researchers towards appropriate performance metrics, software packages and relevant literature links. Mike is also active in the TDP 3b group, which is developing tools and methods to assess
image quality and harmonize data across the MIDRC dataset,
including the use of cohort building tools.
Mike lives in the LA area with his wife, Jill, who is faculty at the University of Southern California, where she directs the Biomechanics Research lab. They have two adult
children who live in the LA area. Together with his wife and children, Mike has attended five different Summer Olympic Games (Atlanta 1996, Sydney 2000, Beijing 2008, London 2012 and Rio de Janeiro 2016); Jill actually did research at the Atlanta and Sydney games as part of the International Olympic Committee’s Medical Commission team and continues to work with Olympic athletes today. While he wasn’t able to attend Tokyo 2020 as they didn’t allow spectators (and yes, sadly, they had tickets!), Mike and his family are looking forward to attending the games in Paris in 2024 and hope to welcome
MIDRC colleagues to LA for the 2028 games!
|
|
|
John Mongan, MD, PhD
John Mongan, MD, PhD, is the Associate Chair for Translational Informatics,
Director of the Center for Intelligent Imaging and an Associate Professor of Clinical Radiology (Abdominal Imaging and Ultrasound section) in the Department of Radiology and Biomedical Imaging at the University of California, San Francisco (UCSF).
He is board certified in both diagnostic radiology and clinical informatics.
John is nationally and internationally recognized as a leader and expert
in artificial intelligence and machine learning. He chairs RSNA’s Machine Learning Steering Committee. Through this committee, he organizes the assembly and curation of large multi-institutional medical imaging datasets and artificial intelligence challenges using these datasets. He serves on the editorial board of the journal Radiology: Artificial Intelligence and represents RSNA in organizing the national AI Safety Summit,
which will be held for the first time in 2022.
His research focuses on artificial intelligence in medical imaging.
He was the senior author and primary investigator on a project that developed artificial intelligence for the detection of pneumothorax. In partnership with General Electric, the algorithm developed in this project achieved FDA clearance and is currently commercially available on portable X-ray machines. He is the lead author on the Checklist for Artificial Intelligence in Medical Imaging (CLAIM), a guideline used by several journals to promote reproducibility in artificial intelligence publications, and is the lead author on a publication drawing lessons for the safe implementation of artificial intelligence
in medicine from the 737 Max disasters.
John believes that data is the lifeblood of machine learning,
and that limited data availability is a primary factor holding back the development of effective AI for radiology. As part of addressing these shortcomings with respect to COVID-19, he was one of the contributors to the RICORD dataset, RSNA’s first foray into COVID-19 data, and has enjoyed continuing to work with MIDRC on technical projects
related to data procurement and ingestion.
John works as an abdominal radiologist at UCSF and lives with his family in San Francisco. He enjoys traveling, and—when he can pry his son away from various electronic devices—hiking and experiencing the outdoors with his family.
|
|
|
David Smith, MD
Through the American College of Radiology® (ACR®), David Smith, MD, has led LSU Health New Orleans to contribute 17,000 images and clinical data from 46,000 patients to Medical Imaging and Data Resource Center (MIDRC). LSU’s dataset greatly enhances the racial and ethnic diversity of MIDRC cases.
Smith’s passion and perseverance in addressing the COVID-19 pandemic
make him an integral member of the CIRR Steering Committee and a true partner in such research. He practices, teaches and lectures in Cardiothoracic, Abdominopelvic and Musculoskeletal subspecialties. Smith lectures nationally and internationally on many topics, including his area of special interest, Interstitial Lung Disease.
When not working, Smith pursues his interests (following years of study)
in Classical Antiquity, language and linguistics, philosophy and logic, Western intellectual history, visual arts and music, and woodworking and stained-glass crafts.
His primary academic appointment is as Associate Professor of Radiology
at LSU Health New Orleans. He also is on faculty at the South Louisiana Veterans Health Care System hospital in New Orleans. Smith obtained his Doctor of Medicine at the Columbia University College of Physicians & Surgeons in New York.
|
|
Our Seminar Series is held on the third
Tuesday of the month from 2-2:45pm CT.
These virtual seminars are free and open to everyone. We hope you'll join us throughout the year for these engaging discussions.
Register here.
|
|
More on MIDRC
Read literature published and peer-reviewed by MIDRC investigators regarding
our research.
We are committed to the crucial principles of equity and inclusion.
Learn more about MIDRC Diversity.
|
|
|
The Medical Imaging Data and Resource Center (MIDRC)
is a collaborative multi-institutional effort between AAPM, ACR and RSNA that is funded by the National Institute of Biomedical Imaging and Bioengineering (NIBIB) and hosted at the University of Chicago to collect and publish medical imaging and related metadata. MIDRC provides a model for collecting and publishing data to support research that can be applied across all areas of medical imaging. The curated images and metadata in the MIDRC Data Commons Portal will support machine learning and Artificial Intelligence research, aiding in the fight not just against COVID-19 but also future pandemics and other pathologies, impacting health outcomes
for the common good.
|
|
|
If you are interested in becoming a MIDRC partner please contact kpizer@bsd.uchicago.edu
|
|
|
|
|
|
