Radiology Rewired, Season 2, Episode 2 | Dr. Vivek Singh with Dr. Dan Sodickson
Most of radiology still runs on a reactive model: a scanner gets used once someone is already sick, to confirm what's already suspected. In Season 2, Episode 2 of Radiology Rewired, Dr. Vivek Singh sits down with Dr. Dan Sodickson, radiologist, physicist, and the researcher whose parallel imaging work helped accelerate MRI during nearly two decades leading innovation at NYU Langone. Now Chief Medical Scientist at Function Health, Sodickson is betting that imaging's next act isn't diagnosis. It's surveillance: tracking change over time, at times, before disease ever declares itself.
Sodickson points to a less-discussed application of AI in imaging: AI that changes how the image gets acquired in the first place. His own parallel imaging work accelerated MRI by capturing data from multiple detectors at once instead of one line at a time, the same principle that makes our eyes fast: many receptors sampling in parallel instead of one point at a time. AI has since pushed reconstruction further, pulling high-quality images from a fraction of the raw data. Scans are already four to five times faster than a decade ago, and when prior imaging exists for a patient, acceleration climbs to 20–30x.
"You can take a 20-minute exam and cut it down to a minute if you have some previous information about that person."
As Sodickson puts it: "There's an upstream use of AI, which is to change the data that we gather in the first place." That shift is what makes population-scale screening plausible: shorter exams and lower dose makes routine, repeated imaging logistically possible for the first time.
That frustration with reactive imaging is what pulled Sodickson out of academia. "It seemed like a crying shame," he says, "because I know how much we can see and characterize disease — so why use our best tools last, rather than first?" He began advising Ezra, an early proactive-imaging startup; when Function Health acquired Ezra, he joined full-time. The model is to track a person's biology and imaging together and continuously, starting long before they're sick. Members begin with 160 blood biomarkers and a baseline MRI of the head, abdomen, and pelvis, less a diagnostic snapshot than what Sodickson calls "GPS for your health," a system that knows where you've been and can flag where you're headed from a health perspective.
"The better we know you, the better we can predict your health, the better we can understand the imaging."
His own research quantifies exactly how much that history is worth: an AI model predicting five-year prostate cancer risk from imaging alone had a 64% false positive rate. Feeding it prior scans plus a small amount of blood data (PSA, age, prostate volume) dropped that to 9% false positive rate.
Dr. Singh, podcast host and practicing radiologist, pressed Sodickson on what this means for the specialty itself. In this format, routine normals may get triaged by AI, but every flagged case will arrive pre-loaded with prior imaging, labs, and genomics. It’d provide insight closer to a consult than a cold read. Sodickson argues: "The job of radiologists could get more intellectually satisfying, not working through 99 normal exams to find the one abnormal, but really using your expertise to figure out what needs to happen."
Singh raised a live example: Rapid Aneurysm-style tracking, which quantifies morphology change across serial scans far more precisely than eyeballing pixels ever could. Sodickson's response cuts to the point: "What we have now is a combination of the best in artificial intelligence, machine intelligence, and human intelligence... we have this whole safety net that hasn't always been there."
Want to hear the full conversation? Listen to Season 2, Episode 2 of Radiology Rewired — available now at [▶ Spotify] [▶ Apple Podcasts] [▶ YouTube] .
Is whole-body MRI accurate enough for routine screening?
It’s a surveillance tool, not a diagnostic exam, designed to establish a baseline and flag the most clinically significant findings. Dr. Sodickson emphasizes that setting patient expectations upfront is essential. Value compounds with serial scanning: more prior context means fewer incidental findings and false positives over time.
How does combining imaging with blood tests improve outcomes?
Adding prior scans and a modest blood panel to an imaging-based AI model reduced false positive rates in prostate cancer risk prediction from 64% to under 10% in Dr. Sodickson’s research. The combination of imaging and biology, updated over time, is where predictive medicine becomes genuinely powerful.
Will proactive imaging always be expensive?
Not necessarily. As scan times shorten, potentially by 20x with AI acceleration, technical costs should fall proportionally. Dr. Sodickson also argues the model will prove cost-effective for insurers once downstream savings from earlier intervention are factored in: a democratization of health that is proactive, personalized, and increasingly accessible.