Can you give a quick overview of your role, especially where it intersects with technology?
I’m an Assistant Professor of Neurology at Rush and Director of AI for Neurology. What that means as it relates to technology is that I function as the point-person for how our neurology department will implement enterprise-level AI solutions, and can also advocate for any specialty-specific solutions that we need. I also help invest in health tech through MIT’s Castor Ventures and as a LP with Scrub Capital.
What key technology areas are you focusing on in neurology?
Image Retrieval Automation: Neurologists frequently need outside imaging (CTs, MRIs), but the current retrieval process is slow and inefficient. We’re piloting a tool that uses LLMs to traverse the health information exchange, locate scans, and upload them directly into our system, reducing wasted clinic visits and unnecessary duplicate tests.
Care Coordination: CMS launched the GUIDE program last summer, which reimburses comprehensive dementia care. Many startups are emerging to support this, and we’re evaluating which solutions best serve our patients.
Stroke Workflow Optimization: We’ve been using Viz.ai for about 18 months to enhance stroke patient triage.
Volumetric MRI Analysis: AI tools that quantify brain changes in MRIs have major research applications and could help in diagnosis and treatment planning.
You’ve been working with Viz.ai for a while now. Are there any generalized learnings you can share from this experience about implementing and working with AI-driven solutions?
The key lesson: Start within your department. If neurology initiates AI adoption, it’s easier for cardiology, pulmonology, or others to follow, rather than trying to gain system-wide buy-in from the start.
You’re also trialing ambient scribing. We’ve heard that quality varies quite a bit depending on specialty and subspecialty—what are you finding?
We’re testing ambient scribes in both general neurology and subspecialties (stroke, epilepsy). The general models work decently for basic notes, but they struggle with subspecialties, particularly when it comes to capturing detailed assessment and plans well. These are the meat of our notes, and so the inability to capture it well, even when I dictate explicitly, can be frustrating.
That said, our overall assessment is that any scribe is better than no scribe. Even if they only handle the history of present illness, that still saves time.
Are there areas where you see AI opportunities but no strong solutions yet?
Many neurology patients have some form of acute disability and require transition to nursing homes, acute rehab, or sub-acute rehab following their hospital stay. So there’s a significant need for post-acute care coordination.
The shortage of case managers and burden of the insurance review process means approvals and transfers take days longer than necessary. AI could help by summarizing clinical data and expediting workflows on both the hospital and payer sides, reducing hospital length of stay. Some companies are working on this, but solutions aren’t fully built out in our experience.
Are there any other areas you’re paying particularly close attention to?
I’m interested in tracking the progression of AI agents from those that can simply execute defined workflows, to those that can operate within a broader scope, to those that can truly function independently and pursue broad, open-ended challenges.
One area I’m watching this is front-office task automation. We need AI that goes beyond basic automation to be able to proactively engage patients. For example, it’s easy for AI to remind patients to book an appointment. But what if they say, "I don’t feel like it" or "I have shoulder pain"? AI should be able to address concerns and guide them toward booking.
There’s huge potential in preventative care and post-discharge follow-ups—reminding patients about screenings, medications, or rehab appointments.
