September 23, 2019—Austin, TX -The Annual Conference on Machine Intelligence in Medical Imaging (C-MIMI) brought researchers, clinicians, and industry experts together to discuss the latest developments in artificial intelligence (AI) for radiology applications. Dozens of scientists presented their research on the latest machine learning algorithms, clinical applications, and datasets during the two-day conference. Radiologists in the audience asked several challenging questions of the researchers, many of which revolved around a central theme:
How can AI solutions provide real value to clinicians today?
This theme was the focus of a heated vendor panel discussion on driving AI adoption in clinical practice. Panel participants, all leaders in the AI vendor space, spent the better part of an hour debating the current barriers, and possible catalysts, for widespread utilization of AI in radiology. Everyone agreed that AI vendors must take their solutions out of the lab and commercialize them to move past the early adoption phase, but this task has proven difficult. The consensus among the panel and the audience was that radiologists do not trust AI solutions to provide value to their practices. As a result, most AI solutions are not commercially viable, but the panel proposed several strategies AI vendors can implement to prove their solutions are good investments.
Develop AI solutions in partnership with radiologists
The major disconnect between the engineers who develop AI solutions and radiologists who use them has led to a mismatch between the AI solutions in development and the problems that radiologists need to solve. The panel thinks AI vendors must recruit radiologists to participate in the earliest stages of algorithm development to bridge this gap. Radiologists should be the ones who define use cases for new AI algorithms, thus infusing this research with domain expertise and ensuring there is market demand (this suggestion garnered applause from the radiologists in the room). When AI solutions are ready for testing, radiologists must be empowered to provide feedback to vendors that will guide the evolution of the products.
Seamlessly integrate AI into workflows
Radiologists in the audience stressed that they are burnt out and stretched for time. They simply do not have the bandwidth to add AI solutions to their workflows. In fact, the most demanded AI solutions are those that increase efficiency and save time. Therefore, AI solutions must be seamlessly integrated with radiologists’ current workflows and quickly deployed to drive adoption. The panel agreed that AI vendors must achieve true interoperability with radiologists’ reporting tools to make their solutions marketable.
Provide valuable, actionable insight
Radiologists desire a diagnostic experience that brings together all the outputs of AI solutions in a way that can be understood and effectively utilized. One frustrated attendee said the AI algorithms he uses often contradict one another and muddle his diagnoses instead of enhancing them. AI vendors must provide radiologists with methods for interpreting algorithm results and incorporating these results into the decision making process.
Focusing on algorithm outputs instead of clinical outcomes is a common mistake among AI vendors. The main takeaway from C-MIMI for these vendors is that they must involve radiologists in every step of the algorithm development process to create valuable, marketable AI solutions. They must also embed these solutions in the radiology workflow while providing radiologists with the tools to make empowered decisions. Increased collaboration between AI vendors and radiologists will speed up the timeline for widespread clinical adoption of AI and usher in a new era of intelligent healthcare delivery.