RSNA 2020 Highlights: Radiologists’ importance in stroke diagnosis and treatment, advances in AI helping reduce imaging radiation, best practices for AI implementation

We wanted to share one more newsletter before the end of 2020. This one is all about the virtual RSNA conference earlier this month. Even if you were one of the almost 30,000 people registered for the conference, it’s hard to attend all of the 350+ live sessions, which included 118 AI sessions. These led to more than 8,000 news media stories written about the conference sessions and research shared. Therefore, we would like to share some RSNA 2020 highlights on AI and neuroradiology. We hope you find this brief recap of select sessions helpful. Please feel free to share it with others.

If you would like to be informed when the next newsletter that we call Cercare Bulletin is out, please sign up here.

Radiologists’ role in diagnosing stroke: A Hot Topic panel session shared how neuroimaging is increasingly playing a role in stroke management. That includes determining whether an acute stroke patient is eligible for revascularization therapy. The panel discussed several trials in this area, with one researcher noting that thrombectomy is used more frequently as a result of the trials’ findings. The panelists discussed the interest in making CT the new minimum standard in stroke work-up, due to several imaging advantages, though MRI still has an important role, especially for assessing targeted therapeutics in assessing stroke patients.

Improving AI radiology performance with federated learning: This session featured work of a multi-institutional team led by a UCLA researcher, to train three algorithms from data sets at separate sites. Researchers found that the models each produced equivalent performances when compared to a standalone algorithm which had been trained on the entire dataset. The UCLA researcher said the study demonstrated that the tool may be ready to use on multi-institutional deep learning models.

E-book

AI in Radiology during Covid-19 and Federated Learning as a New Step for Medical Imaging in 2021

Imaging AI programs to reduce radiation or contrast media doses: Two sessions focused on ways AI can be used in neuroimaging to either reduce scan doses or eliminate the use of gadolinium in certain studies. The no-gadolinium study looked at the feasibility of using deep learning models in primary tumor patient imaging, to generate T1 post-contrast images, by using non-contrast MRI images instead. The model was not able to predict gadolinium enhancement for those with significant motion artifact or in some other cases, though researchers felt that for the latter, incorporating larger datasets in the future could help with training. While more research is needed before introducing this use of deep learning into the clinic, the researchers felt the study demonstrated feasibility. Another presentation showed that deep learning algorithms are being developed that could result in lower scan doses and faster scans, while not degrading the signal-to-noise or contrast-to-noise ratio.

Trust is key to successful radiology AI implementation: Another session discussed the critical elements to implementing AI successfully. Those elements include trust, transparency, and human involvement. This is important given the increasing complexity involved in the medical system when introducing AI. Radiologists need training in how AI works, proper practice integration, understanding how to evaluate AI performance, and background in the ethical issues surrounding AI’s use. Trust in AI usage is influenced by automation bias, the tendency for humans to ignore data that doesn’t fit into the decisions generated by AI or machine-based information.

Conclusion

While the RSNA conference had more than 15,000 chat interactions and a 28% increase in radiologist members attending, we hope to be back in person next year. There’s nothing like the excitement of being able to see colleagues in person and to learn about the various advances in research on site. We were thrilled to see so much interest and research presented on AI and neuroradiology, and look forward to the continuing interest in this field.

Would you like to stay up-to-date with the latest news in AI and neuroradiology? Stay informed with our Cercare Bulletin.