Pietro Nardelli, PhD Instructor in Radiology Brigham and Women’s Hospital…
Aaron Sodickson, MD, PhD and Raphael Cohen: Jump-Starting AI Research in Emergency Radiology: Challenges and Solutions
Aaron Sodickson, MD, PhD Division Chief, Emergency Radiology, Brigham and Women’s Hospital Associate Professor of Radiology, Harvard Medical School |
Raphael Cohen Senior Machine Learning Scientist, Emergency Radiology and Anesthesiology, Perioperative and Pain Medicine |
|
Aaron Sodickson, MD, PhD, FASER is Associate Professor of Radiology at Harvard Medical School. He is the Division Chief of Emergency Radiology at Brigham and Women’s Hospital, and is Past President of ASER, the American Society of Emergency Radiology. His research and administrative efforts include: Emergency Radiology practice management; CT technology assessment and innovation; CT protocol optimization for image quality, radiation dose, and workflow efficiency; Dual Energy CT applications; and Machine Learning in Emergency and Trauma Radiology. | Raphael Cohen is the Senior Machine Learning Scientist in Emergency Radiology and Anesthesiology at Brigham and Women’s Hospital. His research efforts are in the creation of AI-powered detection and mitigation algorithms in both Emergency Radiology and Obstetric Anesthesia. He has also focused on the creation of platforms to facilitate lifecycle AI projects to not only produce effective models, but curate valuable data and insights from raw medical data and imaging. |
Abstract
AI efforts are notoriously difficult to launch in research settings. First attempts can be fraught with challenges, and initial models may be disappointingly unpredictive. Success depends no only on AI techniques and approaches, but on the larger research workflows and processes. We will outline technologies used and developed, supporting processes, and hurdles we faced in starting up AI efforts in the emergency radiology setting. Minimizing friction between workflow processes, collaborating closely between AI expertise and clinical expertise, and adopting and adapting workflow improvements, such as Active ML processes, yielded benefits to the overall research process.