Webinar insights: “How to achieve excellence in Brain MRI Acquisition”

Opportunities abound

“Can AI and related technology positively impact the imaging process”, asked Larry Tanenbaum, MD FACR, CTO and VP of RadNet, setting the stage for the discussion, “having a positive impact on patient experience, supporting and guiding our technologists to produce better exams, and reducing the burden on our radiologists?” In the webinar on “How to Achieve Excellence in Brain MR Image Acquisition,” organized by Cerebriu on November 14, 2023, an expert panel provided valuable insights into the opportunities of current and future technology.

Providing an overview of current challenges and opportunities in MR imaging, in his keynote talk Dr. Tanenbaum focused on the potential of novel AI-based technologies to address issues such as shortage of expert personnel, unsatisfactory patient experience, and frustration among radiologists due to inadequate workflows and processes.

Dr. Tanenbaum envisioned a future with tools like order assistance from large language models based on medical records and indications that can standardize the protocol selection. The MRI vendors are all working to improve user experience with fewer clicks and streamline the acquisition process by deep learning driven acquisition, for example to set ideal angles. This is already improving the exam experience, a statement he underscored with results from his own institution showing a reduction in the time patients spent on the MRI table after deployment of autonomous scanning support.

Another area of improvement is in automated quality assurance of scans, providing technologists with an objective measure to decide whether a scan should be repeated or is sufficient for diagnostic reading. This has the potential to reduce both the number of scans unnecessarily repeated because they were erroneously judged inadequate by the technologist, and the number of patient recalls if the radiologist deems the final result insufficient. Such technologies can assess image quality, noise, motion artifacts, but also appropriate positioning and reformats.

Personalized imaging

“Perhaps the part that we are most interested in today is on-device decision support,” he pointed out. “Imagine a world where something like the Cerebriu software can look at your scans and identify critical findings during the examination, allowing the technologist to either summon someone’s assistance or even guide them to the appropriate protocol.” In some settings this may even affect the standard protocol. Instead of “a shotgun brain MRI” it may be possible to reduce routine scanning to a basic sequence selection and go further with a personalized or disease-based imaging scheme based on AI findings.

Another crucial development to tackle the workforce shortage for Dr. Tanenbaum is the remote operation of MRI scanners by experienced technologists, either giving educational support or running the scanners remotely with a safety trained ancillary personnel on site. “The situation is so bad around the world that scanners can sit idle with no one to run them. It is really a remarkable advance that we now can have people working remotely and running a number of scanners simultaneously,” he concluded.

Prof. Sylvie Destian, Professor Emeritus of SUNY Upstate Medical University, supported the notion that most places have a “shotgun approach to MRI.” She elaborated, “I still feel strongly that we overimage people. We have fewer radiologists, we have fewer technologists, and we need to get the diagnoses with the least amount of imaging.” She particularly expressed her concern about contrast use in children under two.

Also James Hillis, MBBS, PhilD, attending neurologist at Massachusetts General Hospital and Instructor in Neurology at Harvard Medical School, supported the view of reducing imaging to only the appropriate sequences and administering contrast only in cases when it’s really needed. Being driven by the opportunities to improve health outcomes through the use of AI, he sees huge value in this application: if you’ve actually got AI that’s working at the actual scanner console at the same time as the scanner is acquiring these images, then it allows you to have that addon type approach where you can acquire the contrast only for the patients who really need it.”

Workflow, reimbursement and adoption

However, the experts also agreed on the difficult reimbursement setup, particularly in the US outpatient settings, in which such add-on types of imaging scenarios are difficult to be covered. There was general hope from the panelists for change as personalizing and streamlining imaging processes could reduce the number of repeat imaging visits.

On a broader level, Dr. Hillis stated his expectation of more holistic AI solutions from order placement to the final interpretation of that imaging. Michel Nemery, MD, as an innovation leader at RAIT in Denmark, picked up on the point of holistic solutions and presented experience from Herlev and Gentofte hospital, where a pilot explored the routine use of MRI in the emergency department, staffing with neuroradiologists at night and on the weekend. This had increased the percentage of neurological patients in the emergency department to be discharged from 40% to 70%.

Asked about the evolution of MR imaging volumes, in addition to the already rising mismatch of cases to be read and staff to read them, Tanenbaum saw the biggest development in neuroradiology around disease modifying therapies and Alzheimer’s disease, demanding five to ten MR exams per year for as many as 1.5M Americans on these therapies.

In the open discussion about opportunities for AI to alleviate this mismatch, the use case of discerning normal from abnormal exams was controversially discussed but the general sentiment was that AI is not at the stage to make this decision and that radiologists keep this crucial role. However, as Prof. Destian pointed out, alerting the radiologist to abnormal cases can expedite the reporting process. Dr. Nemery raised the opportunity for AI systems to automatically generate quantitative measurements that are utilized in downstream diagnosis. And Dr. Tanenbaum added: “The ability to get patients out of the machine, get the answer right the first time, increase capacity on those machines by making the exam more concise, more consistent, and, frankly, more diagnostic, I think will help.”

On the issue of practical adoption, Dr. Tanenbaum was quite optimistic, stating any of the tools presented would be easy to adopt, at least in terms of the workforce, but pointed out the challenge of paying for such tools. He also gave positive examples: the value of triage has been established quite well, particularly in stroke care and that new workflow-oriented value propositions such as improving reporting to ensure follow-up exams come into focus with large language models. Dr. Hillis also described from an earlier CT-based AI deployment that a lot of value was created by streamlining the processes, in addition to the actual quality improvement by the AI detection. Dr. Nemery added the Danish perspective with its public healthcare system where a holistic assessment of the value for the hospital would be the basis for making investment decisions.

Wish list for future AI support

The closing round revolved around the wish list for future AI support in MR imaging. Tanenbaum asked for any of the tools he had mentioned to be fully integrated into his environment without adding any headwind to his productivity, and also reminded the audience to be aware of important issues such as bias.

Prof. Destian specifically asked for support in volumetric assessment, such as in MS disease burden and generally on the recognition of emergent findings in follow-up scans.

Answering her wish for correct clinical information, Dr. Hillis envisioned a tool to utilize clinic notes and extract the relevant information to populate the MRI order. Tanenbaum put this in the larger perspective of a holistic tool supporting all the way from picking the right protocol to making sure reports are in line with the clinical concern.

Finally, Dr. Nemery focused on the need for vendors to package their algorithms with a validation tool for the local patient population, not just in terms of performance, but also in comparison to the human alternative.

Wrapping up the webinar, many opportunities in the field of AI for excellence in MRI acquisition were pointed out by the experts along with some important challenges ahead for the field. However, what became very clear from the discussion, was that already today many valuable use cases for AI deployment exist to support radiology departments in raising excellence in MRI acquisition.

Cerebriu thanks all participants for their insightful contributions and the open discussion. We are very excited to continue this discussion with more webinars and live events in the future. Reach out to us to engage in the conversation and to understand how Cerebriu can support your institution.