This month’s blog features Tina Shiang, MD, a Musculoskeletal Radiology and Interventions Fellow at Brigham and Women’s Hospital and our 2022 DSI Travel Grant recipient. The DSI Travel Grant covers the costs to attend the Society for Imaging Informatics in Medicine’s (SIIM) Annual Meeting, including the pre-conference ACR Data Science Summit for one ACR member-in-training. 

Imagine the current healthcare landscape: all levels of physicians, nurses, and paraprofessionals work tirelessly to keep up with the ever-growing clinical workload and keep our patients safe. Imaging volumes continue to increase rapidly. Radiologists are being relentlessly challenged to accurately interpret an exponential number of images while also dealing with numerous non-clinical tasks, amid staffing shortages and burnout. Our system is dysfunctional, and we are struggling. We need to do better, and in recent years, we have come to realize that maybe we can. There has been a surge of interest in artificial intelligence (AI), and our investment in healthcare informatics has already begun to transform medicine and unlock new opportunities. 

ChatGPT and Large Language Models in Medical Imaging 

Since its release in November 2022, OpenAI’s ChatGPT (Chat Generative Pre-training Transformer) became the most popular AI chatbot, generating a lot of excitement and public interest. ChatGPT is a large language model (LLM) that uses transformer technology for natural language processing and can generate human-like responses. Historically, AI adoption in clinical practice has been low and slow. The 2020 ACR Data Science Institute Artificial Intelligence Survey published in JAMA found that only 30% of radiologists use AI in clinical practice. By a show of hands at this year’s SIIM conference, still few radiologists in attendance had experience using AI in clinical practice, yet almost all had tried ChatGPT. For those who had not yet had the opportunity, the SIIM Annual Meeting offered its inaugural AI Open Source Playground, where attendees could interact with AI in real-time for an immersive and practical experience.

At this year’s ACR Data Science Summit, Dr. Keith Dreyer from Mass General Brigham spoke of the “triple transformation” in radiology from modality to digital, and now to include AI. He shared that he estimates we are hovering at approximately 2% penetration of AI in healthcare but is hopeful that ChatGPT would be the catalyst that pushes more adoption of AI in radiology. Much of the discussion involving ChatGPT centered on how to leverage it effectively in the clinical environment. There are many promising uses including streamlining workflow, reporting and communication, and error detection and reduction, just to name a few. These topics led to a broader discussion on multimodal models, which expands the contextual scope to capturing relationships between different data modalities (i.e. text, images, videos, audio) and making predictions based on a comprehensive understanding of mixed data types. 

LLMs (such as ChatGPT) and multimodal models are gamechangers, but there is also controversy. The models may produce factual inaccuracies or unpredictable biases that may worsen health inequities in vulnerable populations, and we must be aware of these limitations to prevent unintentional harm to patients. Many of these models attempt to replicate human behavior, often with varying degrees of success due to human inconsistencies. In a keynote presented by Dr. Ziad Obermeyer from UC Berkeley, Dr. Obermeyer felt that we should aim higher, and AI should do better than humans. For example, these may include applications in opportunistic imaging and digital biomarker analysis for precision medicine, where AI may provide potential insights that may be imperceptible to the human eye. When “AI can do things that humans cannot do,” and augment the radiologist, that is “value-added.” 

Value-Based Radiology, Return on Investments, and AI Governance 

Healthcare has moved towards a value-based model, and radiologists must provide high-value imaging that is both efficient, cost-effective, and comprehensive. In the words of Dr. Richard Bruce from the University of Wisconsin-Madison, we must “establish an AI value proposition because there is a finite budget and resources in radiology for competing interests.” AI has great potential, but a great idea does not necessarily equate to successful implementation and longevity. In fact, the “complexity in implementing and maintaining AI solutions” is often more challenging than the algorithm “design” itself. 

Dr. Howard Chen from the Cleveland Clinic proposes a way to begin to address this following SMART criterion by setting specific, measurable, achievable, relevant and time-bound goals. He and other speakers outlined the importance of establishing local AI governance and going through the process of methodically and clearly defining the clinical need and problem you aim to solve, identifying the target audience and stakeholders, justifying the expected value and return on investment, and developing a robust framework for implementation and monitoring. The process of aligning value with incentives is not easy, but “you have to start somewhere” and persistence is key! 

Building a data science team may help build momentum to overcome bureaucratic inertia. Panelists Dr. Paul Yi and Dr. Vishwa Parekh from University of Maryland Medical Intelligent Imaging Center, Dr. Veronica Rotemberg from Memorial Sloan Kettering, and Dr. Walter Wiggins from Duke University, gave the audience useful pieces of advice from their own institutional experiences: 

1. A recurrent theme was that “AI is a team sport.” 
2. Collaboration should be multidisciplinary and include all stakeholders (including clinical champions, technical specialists, decision makers and industry sponsors, etc.) in a shared decision-making process. 
3. Finding the right team members takes time and “the best team members are those who are motivated and whose interests align with the project vision and goals, not necessarily those with the broadest skill set.” 

Building Relationships and Promoting AI Innovation 

The COVID-19 pandemic necessitated social distancing. This helped drive the rise of widespread use of virtual communications for safer social interactions, but also exacerbated social isolation as it limited in-person connections. Dr. James Whitfill from University of Arizona emphasized that “humans need social connections that cannot be fully replaced with virtual tools” and that “being together within a culture of belonging is what makes SIIM so unique.” Even as a first-time attendee at SIIM, I can appreciate the experience he describes. 

I met trainee colleagues from diverse backgrounds and from all around the world at the SIIM Members in Training Meetup. I even reconnected with a friend I had met 5 years ago at RSNA as a first-year radiology resident, both of us delighted to discover that we shared another common interest beyond waakye and jollof rice. I joined our more adventurous colleagues on the SIIM23 5k Fun Run/Walk on a toasty, beautiful Austin summer morning to challenge our physical tenacity – my unfortunate directional sense taking me on a longer scenic detour along the Colorado river, where I met many friendly Austinites. 

SIIM also hosted many networking opportunities. There was friendly competition at the Real-World Rochambeau Rock, Paper, Scissors competition and by chance, I met this year’s champion at the networking lounge. We were able to meet the luminaries in informatics and chat over food and drink at events such as the SIIMfund Silent Auction/Happy Hour. For more focused discussions, Dr. Tessa Cook from Penn Medicine guided us through several small group round tables on professional development through an informatics career, focusing on mentorships and sponsorships. We shared personal experiences on some of the challenges women in informatics face, such as imposter syndrome and setting personal and professional boundaries, and ways to address them. Teri Sippel Schmidt from Johns Hopkins and Marquette University, our closing keynote speaker, told her inspirational story and journey through the world of science and informatics, leaving us with an important message: “one role model can make all the difference in the world” to the young minds who will create the future of the world and we must all reach out to support and foster their talent. 

Learning, Innovating, Belonging

SIIM has truly been a special experience for me as an early career physician-informaticist, and will no doubt remain a memorable one even many years down the line. If, as a physician/radiologist, you have felt uncertain or intimidated by AI, I hope my experience can convince you to attend SIIM next year and take the first step in exploring the world of informatics.

Radiology is at the intersection of healthcare and technology. We can transform how we practice medicine and deliver care if we embrace our role as AI-enabled radiologists. Together, with the help of our multidisciplinary colleagues, we can move innovation forward and create a healthier future for our patients.

In the spirit of embracing new technologies, I leave you with some parting words from ChatGPT and this year’s phenomenal conference: 

At SIIM’s meeting in ’23, 

Learning, innovating, belonging we see. 

A gathering of minds, sharing their art, 

Advancing healthcare with a united heart. 

Tina Shiang, MD | Musculoskeletal Radiology and Interventions Fellow at Brigham and Women’s Hospital | Informatics Fellow at the Center for Evidence Based Imaging