The American College of Radiology Data Science Institute has released its first-of-a-kind series of standardized artificial intelligence use cases to speed up the adoption of quality AI technology in medical imaging.

The initiative is to form a framework of guidelines on the functional aspects required in an AI solution so that developers can create effective algorithms that can be properly implemented in clinical practice and that clinicians can use to establish quality test models and analyses for collecting the information they need.

“The DSI is works with clinical experts from different specialties, industry and data scientists to create standardized use cases for developing AI algorithms that will solve the most important clinical problems radiologists encounter,” Bibb Allen, chief medical officer of the ACR Data Science Institute, told HCB News. “Compared to algorithms developed by industry developers or at single academic institutions, AI solutions that follow DSI AI use case standards will be able to be readily integrated into clinical practice across multiple workflow systems.”

Deemed the ACR-DSI Technology Oriented Use Cases in Health Care-AI (TOUCH-AI) framework, the basis of the guidelines is for algorithms to address relevant clinical questions; be capable of being implemented across multiple electronic workflow systems; enable ongoing quality assessment; and meet legal, regulatory and ethical requirements.

To achieve such objectives, the TOUCH-AI Framework enables multispecialty and multi-industry expert panels to flock together and determine clinically relevant use cases for the development of medical imaging, interventional radiology and radiation oncology AI programs. It also creates a methodology of tools and metrics for creating algorithm training, testing and validation of data sets around use cases, along with standardized pathways for implementing AI algorithms in clinical practice.

With its implementation, users will have opportunities to monitor the effectiveness of AI algorithms in clinical practice through mediums such as the ACR National Radiology Data Registry or the ACR DSI algorithm monitoring service, Assess-AI.

Using these registries provides access to data elements in structured uses cases for the capture of real world performance data that can be utilized for ongoing quality improvement and to help developers comply with legal, regulatory and ethical requirements for medical imaging, interventional radiology, and radiation oncology AI programs.

Allen says the aim of DSI structured use cases is to bring together multiple institutions to create data sets for a particular use case that can be trained and tested with more technical, geographic and patient diversity than those created at single institutions, thereby reducing algorithm bias.

“The next step is for medical imaging algorithm developers and data scientists to become familiar with the AC DSI use cases and follow this method so that algorithms can move beyond being intellectual curiosities that pop up in medical journals or the press and deliver on their promise to safely and effectively improve patient care,” he said. “There is no other large scale framework that can move promising medical imaging AI ideas to effective clinical practice tools so safely and efficiently.”

ACR DSI announced its intention to build a use cases framework last November at RSNA for the first time as part of its objective to convert artificial intelligence from a concept into an everyday practice in radiology. To complete its endeavor, the association teamed up with partners, including the Medical Image Computing and Computer Assistance Intervention Society (MICCAI).

Its first uses cases were released in July for feedback.