The radiology artificial intelligence (AI) environment is growing rapidly, with an increase in product offerings to support efforts in image acquisition, reading, triage and clinical decision support. A sustained increase in imaging orders and the availability of high-resolution scanners have also led to large workloads for radiologists[i],[ii]. However, lower resolution imaging, such as X-Ray, remains one of the most common modalities relied upon for initial diagnosis.
Chest X-rays are performed in 34.4 percent of Emergency Department (ED) visits[iii]. Despite their widespread use, however, X-rays are not the easiest to interpret. AI developers are focused on creating AI solutions to help radiologists read X-Ray studies more accurately.
To evaluate some of the progress AI is currently making in X-Ray, as well as what may lie ahead, a recent webinar, “AI and the Future of X-Ray,” was hosted by GE Healthcare. The webinar discussion featured prominent clinicians from the University of California, San Francisco (UCSF) who shared their experiences in the development and use of AI solutions in X-ray. Using AI tools is helping to automate some manual tasks for the technologist and helps provide reading physicians with clinical decision support. To help prioritize critical cases, AI tools are being used to help detect subtle or complex patterns within X-ray images, moving critical cases to the top of the reading list. Using these AI tools has resulted in improvements across the board in efficiency, quality, and clinical accuracy.
Moving from concept to reality in AI solutions for X-ray
With more than 100 AI software solutions created for healthcare from more than 55 vendors[iv], more and more AI algorithms are moving from concept to reality, proving their value, earning approvals from regulatory entities, and being introduced into everyday clinical practice.
John Mongan, MD, PhD, Vice Chair for Informatics and an Associate Professor of Clinical Radiology (Abdominal Imaging and Ultrasound section) in the Department of Radiology and Biomedical Imaging at UCSF, shared some of his early experiences with AI in radiology.
“From my very first project, working on deep learning AI several years ago,” Dr. Mongan explained, “it was clear to me that this wasn’t just a stepwise, incremental improvement from where we were with computer vision. AI was really something revolutionary. My goal is harnessing AI as the newest modality in radiology that can enable us to perform radiology more efficiently, more effectively and have richer diagnoses.”
Dr. Mongan was involved in the initial development and is currently using GE Healthcare’s Critical Care Suite* to improve quality, safety, and efficiency of care for patients at UCSF.
“Dr. Mongan and his colleagues at UCSF have been involved from the initial concept development of Critical Care Suite all the way through to evaluation and beta testing of the latest version of our AI solution,” noted Alec Baenen, X-ray Product Manager in Artificial Intelligence & Analytics at GE Healthcare. “In order to create meaningful AI solutions like Critical Care Suite,” he added, “it’s important to have that kind of close collaboration with your clinical partners.”
Seamless integration of AI in clinical use
Kimberly Kallianos, MD, Assistant Professor and Director of the Cardiac & Pulmonary Imaging Fellowship Program at UCSF stressed that the value of any AI application might differ depending on needs in each clinical setting. For instance, some areas might have less access to skilled practitioners to read radiology images, so certain types of AI solutions might be incredibly useful to them. In a larger hospital system like UCSF, Dr. Kallianos said she was most interested in using AI solutions to highlight the most abnormal images and bring them to radiologists’ attention quickly, as well as using AI to improve the efficiency in the radiology workflow. She presented three patient case studies highlighting the added value AI-embedded solutions brought to each situation and patient’s care.
GE Healthcare’s Critical Care Suite is a collection of AI algorithms embedded on X-Ray systems for automated measurements, case prioritization and quality control. It features tools intended to assist technologists to correct anatomical positioning and confirm protocol selections at the time of the scan, and auto rotate images to save time for reading radiologists. To improve triage of urgent cases, the AI also automatically analyzes images, upon acquisition, for critical findings such as pneumothorax. Triage notifications are sent directly to PACS and flagged for prioritized radiologist review.
The newly added feature in the AI that illustrates the position of the endotracheal tube and its distance from the carina is one that has also been helpful for Dr. Kallianos. Using this feature, she has been able to monitor patients and alert clinicians when an endotracheal tube advances into a non-optimal position and needs to be corrected.
In one of the cases she presented, Dr. Kallianos highlighted a patient with a trace, right-sided pneumothorax. She demonstrated the AI notation of a suspicious finding for pneumothorax, as well as the overlay which illustrated endotracheal tube localization and provided a distance measurement from the carina. This is a case Dr. Kallianos used for training purposes. The student with whom she was reviewing the case did not see the pneumothorax in the original image, but with the secondary image and the AI annotated captures, this became a great teaching case for her.
Next steps for AI in X-ray
The UCSF physicians discussed their visions for the future of AI in X-ray as well as its potential applications across healthcare.
“We’ve seen all kinds of amazing things in computer labs, product demonstrations and research papers about the potential for AI,” said Dr. Mongan. “The next major challenge will be that transitional step of getting it to more clinicians and into their clinical workflows. If we can bring AI into the information environment so that clinicians can have contextual information, such as a patient’s history of pneumothorax, this will be of great value for radiologists. If there is a suspected pneumothorax, for example, it would be of great clinical importance for radiologists to know whether or not it was new. This is where we need to get to.”
Applying AI results longitudinally to allow clinicians to review a patient’s history over time is one key area GE Healthcare continues to work on, according to Baenen. Across clinical specialties and modalities, the availability and adoption of AI applications is quickly growing, and impacts are evident in workflow improvements, as well as improvements in triage, diagnostics and patient management. AI based tools are becoming a necessity to alleviate the time demands on radiology workflows, as well as provide clinical decision support to reading physicians. Taking the time to develop the AI tools that serve the needs of clinicians is a key factor in the future of AI in healthcare.
* Critical Care Suite 2.0 is not cleared or approved by the FDA. Distributed in accordance with FDA imaging guidance regarding COVID-19 public health emergency.
Not all products or features are available in all geographies. Check with your local GE Healthcare representative for availability in your country.
[i] Zha N, Patlas MN DR. Radiologist burnout is not just isolated to the United States: perspectives from Canada. J Am Coll Radiol. 2019;16(1):121-123.
[ii] Kane L. Medscape National Physician Burnout, Depression & Suicide Report 2019. medscape com/slideshow/2019-lifestyle-burnout-depression-6011056. Published online 2019.
[iii] Fatihoglu E, Aydin S, Gokharman FD, Ece B, Kosar PN. X-ray Use in Chest Imaging in Emergency Department on the Basis of Cost and Effectiveness. Acad Radiol. 2016 Oct;23(10):1239-45. doi: 10.1016/j.acra.2016.05.008. Epub 2016 Jul 15. PMID: 27426978.
[iv] Healthcare Dive January 2020 (from GE ppt).