Radiology departments can adopt responsible frameworks to deploy AI

Adopting a responsible framework could help overcome challenges in applying AI to medical imaging, according to an article published in the Canadian Journal of Cardiology.

Researchers led by MD/PhD candidate Alexis Nolin-Lapalme from the University of Montreal described an open-source software program that they developed, integrating AI models into PACS, called PACS-AI. This approach aims to increase the evaluation of AI models by making their integration and validation easier with existing medical imaging databases.

"It's like the training of a team. The training of an algorithm needs to be well understood," Nolin-Lapalme told AuntMinnie.com. "It's like any clinical tool as well. We need to understand when to use it and how to interpret the results."

As AI continues to surge in popularity among radiology departments, some departments may face unique challenges in applying the technology to medical images. These hurdles include heterogeneity among healthcare system applications, reliance on proprietary closed-source software, and rising cybersecurity threats.

The researchers also highlighted that before AI models are deployed in clinical settings, they must show their effectiveness across a wide range of scenarios and be validated by prospective studies. They added that using AI techniques in healthcare raises significant legal and ethical issues.

Nolin-Lapalme said that a responsible framework is making tools that will perform as equally as possible between medical groups. This includes training AI models that understand or consider underlying biases when dealing with various patient populations.

"I think there's a lot of interest in AI. People think that the performance seems very high," Nolin-Lapalme said. "Sometimes, the results can seem plausible, but understanding with criticism is key."

 

The researchers in their review described PACS-AI, an open-source, vendor-agnostic software application that aims to increase the evaluation of AI models by making their integration and validation easier with existing medical imaging databases. The goal is to offer a pathway toward responsible, fair, and effective deployment of AI models in healthcare, the team wrote.

The platform works as an interface between existing clinical PACS and AI models. The researchers highlighted that its main goal is to allow for automated, near real-time application of AI models on clinical images for use at the point of care.

The platform offers a web application interface that clinicians can use to search for an imaging study stored on the hospital PACS and select a compatible AI model to be applied to the associated images. The application backend then collects the relevant images, prepares the data, and calls for an AI inference to be performed. The user is then presented with the results in the web interface.

Currently, the PACS-AI platform is being used for research purposes only across several Canadian and U.S. hospitals. The authors wrote that once the models are validated and have regulatory clearance, PACS-AI will also be used for the deployment of clinical AI models. They also noted that regulatory clearance in both countries will be needed for clinical use.

The full description of PACS-AI can be found here.

Page 1 of 368
Next Page