3 Shocking Challenges Stopping AI from Taking Over Radiology! (But 1 Has a Solution!)

CONTENTS: 3 Shocking Challenges Stopping AI from Taking Over Radiology!

Researchers, led by MD/PhD candidate Alexis Nolin-Lapalme from the University of Montreal, have proposed a responsible framework to address challenges in applying AI to medical imaging. Published in the Canadian Journal of Cardiology, their study highlights an open-source software program called PACS-AI, which integrates AI models into Picture Archiving and Communication Systems (PACS).

This software aims to facilitate the evaluation, integration, and validation of AI models using existing medical imaging databases. Nolin-Lapalme emphasized that training an algorithm is akin to training a team; it’s crucial to understand when and how to use it, as well as how to interpret the results, similar to any clinical tool.

As AI gains traction in radiology departments, these departments may encounter specific challenges when applying the technology to medical images. These challenges include the heterogeneity among healthcare system applications, dependence on proprietary closed-source software, and increasing cybersecurity threats.

The researchers emphasized that AI models must demonstrate their effectiveness across diverse scenarios and undergo validation through prospective studies before being deployed in clinical settings. They also pointed out the significant legal and ethical issues that arise when using AI techniques in healthcare.

Nolin-Lapalme noted that a responsible framework involves creating tools that perform consistently across different medical groups. This includes training AI models to understand and account for underlying biases when dealing with various patient populations.

3 Shocking Challenges Stopping AI from Taking Over Radiology! Nolin-Lapalme noted the widespread interest in AI, highlighting that while the performance of AI models may appear high, it is crucial to understand the results critically. In their review, the researchers described PACS-AI, an open-source, vendor-agnostic software designed to facilitate the integration and validation of AI models with existing medical imaging databases. The aim is to ensure the responsible, fair, and effective deployment of AI models in healthcare.

PACS-AI functions as an interface between existing clinical PACS and AI models, enabling automated, near real-time application of AI models on clinical images at the point of care. The platform provides a web application interface that allows clinicians to search for imaging studies stored on the hospital PACS, select a compatible AI model, and apply it to the images. The backend of the application then gathers the relevant images, prepares the data, and performs an AI inference, presenting the results to the user through the web interface.

At present, the PACS-AI platform is used solely for research across several Canadian and U.S. hospitals. The authors stated that once the AI models are validated and receive regulatory clearance, PACS-AI will also be used for deploying clinical AI models. They emphasized that regulatory approval in both countries is necessary for the platform’s clinical application.