Cercare Bulletin, Issue 5 from August 5, 2020

ECR 2020 highlights; HIMSS moved to August 2021; AI’s impact in multiple sclerosis & Alzheimer’s disease

Welcome to our fourth monthly newsletter on neuroradiology and AI. We hope you find it helpful. Please feel free to share it with others.

If you would like to be informed when the new issue is out, please sign up here.

ECR 2020: Challenges in AI data use for using clinical practice, and in data scientist/radiologist collaboration

At the European College of Radiology (ECR) 2020 virtual meeting, Nickolas Papanikolaou, Ph.D., head of Computational Clinical Imaging Group at the Center for the Unknown at Champalimaud Foundation in Lisbon, Portugal, discussed AI’s current use in medical imaging and its clinical integration challenges. For AI’s use in imaging, a big draw is that growth in imaging studies outpaces the number of trained radiologists available to read them. The industry needs more external validation for AI models to reduce bias and increase transparency and understanding, he said. Presenter Ben Glocker, Ph.D., who works on machine learning for imaging at Imperial College in London, said that radiologists must consider data quality, variety, volume and readiness when creating, training, or validating models for study, and must ask critical questions when reviewing existing studies. Other challenges are that large volumes of data are required for proper AI training, the regulatory process is difficult, and developers must collaborate with information technology vendors to integrate algorithms into the radiology workflow.

In a separate presentation, Luis Martí-Bonmatí, Ph.D., director of medical imaging and chairman of the radiology department at La Fe University and Polytechnic Hospital in Valencia, Spain said that radiologists need to collaborate more with data scientists to improve AI algorithms and expand their use in clinical practice. He said clinical adoption of AI is hampered by algorithms focusing on narrow tasks rather than end-to-end solutions linking together niche AI algorithms. He said that creating ensembles of neural networks will improve radiology practice and outcomes in the real-world environment.

CAPILLARY DYSFUNCTION AS A COMMON DISEASE MECHANISM

ON-DEMAND WEBINAR

Neuroradiology and AI in the news

Current status and future directions for artificial intelligence in neuroradiology. This article in the American Journal of Neuroradiology states that “There is growing momentum behind leveraging artificial intelligence techniques to improve workflow and diagnosis and treatment and to enhance the value of quantitative imaging techniques.” It also explains why neuroradiologists should get involved with AI and pay attention to its usage. Authors note that AI advances for radiology include: disease detection, lesion quantification, image reconstruction, segmentation, outcome prediction, workflow optimization, and scientific discovery. Interest in the area is growing exponentially. Oral abstracts, posters and AI exhibits at the American Society of Neuroradiology (ASNR) meetings quadrupled from 19 in 2018 to 80 in 2020. Authors argue that neuroimaging experts should be involved with developing AI applications alongside vascular specialists, and that AI is an historic opportunity to help lead and drive positive changes in clinical practice.

AI advances in multiple sclerosis imaging. This review article in the German publication Fortschr Röntgenstr provides an overview of imaging and AI applications for multiple sclerosis, including optimization of lesion visualization and segmentation, synthetic imaging (which is generated by a computer algorithm, not data measured during an MRI scan; used to optimize contrast between multiple sclerosis lesions and brain parenchyma), and predicting clinical endpoints. It concludes that AI is a hot topic in multiple sclerosis imaging, and applications already exist for use in clinical practice. The AI focus for the future in multiple sclerosis is to improve the ability to predict disease course and optimize individual treatment decisions.

Applications of machine learning to diagnose and treat neurodegenerative diseases. The routine use of AI technologies in clinical practice can be for faster detection of neurodegenerative disorders like Alzheimer’s and Parkinson’s disease, leading to earlier use of disease-modifying treatment. This is according to a review from University of Sheffield’s Neuroscience Institute, published in Nature Reviews Neurology. The study investigated using AI to identify the best course of treatment based on a person’s disease progression, predicting which patients with current mild cognitive impairment will develop Alzheimer’s disease, predicting the severity of loss of motor skills over time, and identifying methods to find new drugs and therapeutic targets. Researchers advised that AI can be used to recognize disease-causing changes not only in imaging, but in speech or video recordings of the patient’s voice or movements, reducing the need for in-person visits.

Federated learning in AI: Multi-institutional collaborations without sharing patient data. Deep learning is important in identifying complex patterns, but individual institutions may not have large or diverse enough datasets to train the algorithms. Privacy issues block institutions from sharing patient data. This study from University of Pennsylvania, published in Scientific Reports, discusses the federated learning approach for multi-institutional collaborations. The federated model is able to use data from each institution without sharing it, by distributing the model training to the data owners, and aggregating results. The authors found 99% of model quality achieved, and they could evaluate generalizability of data from organizations outside of the federation. They anticipate that the federated learning approach will be adopted, allowing for training based on large data sets, and improving precision medicine in the future.

HIMSS moved to August 2021. The Healthcare Information and Management Systems Society (HIMSS) moved its March 2021 conference to August 9-13, 2021 in Las Vegas, Nevada. The March 2020 conference in Orlando, Florida was cancelled due to COVID-19.

Conclusion

Research into AI algorithms for neuroradiology usage continues to find both clinical need and applicability. Radiologists and neuroradiologists are encouraged to learn as much as possible about how AI can improve efficiency and effectiveness of diagnoses, bringing higher quality and more timely care to patients with acute or developing chronic needs. Partnering with data scientists and vascular specialists as well as other institutions to develop and test algorithms brings a greater depth to them, and applying approved algorithms for patient neurological imaging can improve patient care.

Would you like to learn more about how AI works and how it can be implemented in your department? Reach out to our experts