Dr. Seong Mun
Arlington Innovation Center of Virginia Tech in Arlington, VA, USA
Seong K. Mun, PhD, is professor and director of Arlington Innovation Center of Virginia Tech in Arlington, Virginia. He has been engaged in research and development activities in computer aided diagnosis (CAD) in diagnostic imagining over the past 30 years. He is also the President and CEO of Open Source Electronic Health Record Alliance (OSEHRA), which was established in 2011 by the US government. It is a not-for-profit organization promoting open source software strategy for community based rapid innovation in health IT for the global market. He is working to develop a global consortium for medical imaging AI based on the open source strategy of open collaboration.
During the 80’s and 90’s Dr. Mun at Georgetown University Medical Center as Director of Imaging and Information Systems Center, he championed the development of filmless radiology (a.k.a. PACS), global teleradiology, telemedicine, e-health health informatics, medical robotics, combat casualty care, and global disease surveillance.
Prior to Georgetown, he was a radiology research faculty at the Columbia University Medical Center in New York where his team built one of the first high field MRI systems in the world, a project sponsored by Philips Medical Systems. He received General Thurman Award for Excellence in Advanced Military Medical Technology from the US Army.
His educational training includes PhD in physics from State University of New York, fellowship at the Department of Radiology, University of Colorado Medical School and BS in Physics from the University of California, Riverside, CA.
Session 6: AI/DEEP LEARNING IN MEDICINE & NLP
DAY 3: September 13, 2019 | Session 6 | 3:35 PM – 3:55 PM
Artificial Intelligence in Radiological Imaging: Lesson Learned and Possible Roadmap Ahead
Seong Mun, PhD, Virginia Tech, Virginia, USA
Over the past 30 years, the radiology community has been developing computer aided diagnosis (CADx) capabilities using convolution neural network before the artificial intelligence became popular. However, AI in imaging has yet to make significant global improvements in radiology while some are concerned that AI might replace radiologists. It is generally recognized that the full potential of AI in imaging and informatics is yet to be realized. This proposed presentation is intended to address the following remaining issues:
- The current machine learning algorithms developed in CNN tools (Tensorflow etc.) are based on recognition of alphanumeric handwriting and general images. However, Radiology images have fundamental differences from non-medical images such as the requirement of perceiving subtle gray value features within a local area.
- To build a robust CNN based tool, a massive and clinically representative data base with ground truths must be established for extensive training and validation. The cost of such a curated data base containing normal and a range of abnormalities associated with a disease can be prohibitively expensive. What and how should the curated data sets be developed research?
- The CNN tools that work in a R&D environment do not easily translate into the clinical setting. Many other issues such as local practice patterns and workflow can impact the effective use of CNN tools.
The paper will conclude with a brief glimpse of what radiology might look like in a foreseeable future.