The Evolution of AI in Enterprise Imaging
Enterprise imaging is undergoing rapid evolution, especially as artificial intelligence (AI) transforms analytics. AI has the potential to revolutionize the way vast amounts of imaging data are integrated and utilized within healthcare organizations, with the power to automate image review and reporting, streamline workflows, and enhance diagnostic accuracy.
The Latest Trends in AI: Generative, Agentic and Interpretive
Burnout remains a significant challenge in medicine. While contributing factors can vary within specialties, common ones include excessive workload, workflow issues, and too much time spent on documentation, and reporting.
Sonia Gupta, MD, is Chief Medical Officer of Enterprise Imaging at Optum, a company dedicated to turning real-world data into valuable, organizational insights. She noted that AI investment has shifted away from tools that assist diagnosis to those that can handle administrative tasks and non-clinical care. “Physicians want to stay focused on patient care,” Dr Gupta explains, and advancements in generative, agentic and interpretive AI are making that possible.
Woojin Kim, MD, Chief Medical Officer of the American College of Radiology Data Science Institute® (ACR DSI), has seen an explosion in generative AI, which can draft reports automatically from images while increasing precision and accuracy. In addition to saving time, the technology mitigates reporting errors that reflect poorly on radiologists. “Our patients don't know what we look like. They only know and judge us by our radiology reports,” Dr. Kim explains.
Agentic AI refers to artificial intelligence systems that can operate autonomously. Without intervention, agentic AI can help manage workflows around patient care activities (e.g., appointments, referrals, test results), clinical decision support (by ordering tests and initiating diagnostic pathways), and resource allocation.
On the analytical level of operations, interpretive AI helps interpret or analyze complex models, data, or results in a way that makes them more understandable, relevant, and applicable to specific contexts or populations. In healthcare, interpretive AI can take large, general AI models and adapt or refine them using local or specific patient data, ensuring the models are relevant to a specific patient group or environment. Combined, these technologies are already helping to address some of the key factors associated with burnout among physicians, while also improving processes and business models that affect patient care.
Real-World Uses of AI and Data Management
Each year, radiologists around the world review billions of medical images. This imaging data constitutes 90% of all healthcare data, underlining its importance in diagnoses and treatments for millions of patients across populations.1 “Storing and gathering insights from that data is a great opportunity for population health. I’m looking forward to seeing how we can help patients [by] gathering trends from specific risk factors,” said Dr Gupta.
In addition to population health, access to data has a significant impact on radiologists’ workflow and well-being. Optum leverages cloud imaging technology to streamline healthcare data storage, access, and analysis, improving efficiency and patient care. The company partners with Google Cloud to develop and offer specialized, cloud-based solutions for healthcare systems.
“Moving to the cloud is about being able to look at images efficiently, especially if you have a surgical team waiting on you or you need to make decisions based on these images,” said Dr. Gupta.
Dr Gupta recommends speaking to colleagues who have already moved to the cloud to learn from their experience, then identifying stakeholders that will play a part in this transition. In this process, it’s critical to determine resource planning and governance structure. “It’s a team effort and they all need to be on the same page,” she said.
However, this transition is a process that requires patience and a plan. “Moving data can take longer than we expect. Resource planning for the hospital system and getting everyone on board also takes time. That’s why some organizations choose to go hybrid, moving data to the cloud gradually,” she said.
Dr Kim is a proponent of moving data to the cloud to speed image loading and streamline the image review process. “As long as you have internet, you can access data anywhere. That enhances collaboration through seamless sharing of data between users. And when it comes to cloud computing, it provides the scalable storage and computing power that we need,” he said.
Generated AI technologies like large-language models and other foundation models are well-suited to being stored in the cloud because of their increased computing needs. “If you want to maintain various solutions, it can be cost prohibitive to do that with on-prem servers. Cloud computing democratizes access for everyone involved,” he said.
As the entire healthcare industry adapts to the changing world of AI, partners like Optum and the ACR DSI will be critical to driving innovation, facilitating implementation, and delivering transformative solutions that improve patient outcomes and operational efficiency. Dr Kim encourages radiologists and facilities to tap into resources from the ACR DSI, which facilitates the development and implementation of AI applications to help improve patient care.
The latest AI solutions go a long way towards alleviating burnout through by reducing mundane tasks, but there is room for improvement. For example, they detect and measure structures and lesions, yet radiologists must manually dictate those measurements into their reports. Dr Kim believes the next generation of AI must address this, saying, “When AI measurements are automatically inserted into the radiology reports, that allows radiologists to practice at the top of their license.
REFERENCES
- Zhou SK, Greenspan H, Davatzikos C, et al. A Review of Deep Learning in Medical Imaging: Imaging Traits, Technology Trends, Case Studies With Progress Highlights, and Future Promises. Proceedings of the IEEE. 2021;109(5):1-19. doi:https://doi.org/10.1109/jproc.2021.3054390