The first cloud-native AI radiology platform. It does the busywork - measuring, segmenting, drafting - so radiologists can do the reading.
No VPN. No three-minute login to a desktop client built in 2009. No second monitor running the worklist while a third runs the dictation software that mishears "no acute findings" as "no cute fillings."
There is just a tab. Inside it: the scan loads in under a second, the worklist already knows which study is a STAT, and a draft report - written in this particular doctor's own phrasing - is waiting to be read, edited, and signed. The radiologist looks at the image. That is the whole job now. That is the entire pitch of New Lantern, and it is a surprisingly radical one.
New Lantern calls the end state an "AI radiology resident" - software that handles the grunt work a junior trainee would, without needing coffee or a weekend. The company is small, about 29 people in San Francisco. The ambition is not.
"Radiologists are already great at finding diseases. What they hate is the boring stuff."
- Shiva Suri, founder & CEOFor decades the story about AI and radiology was a horror story for radiologists. In 2016 a Nobel laureate suggested hospitals should stop training them; the machines would read the scans. The machines, it turned out, were not nearly so eager.
Because finding a tumor was never where radiologists spent their day. They spent it drawing measurements by hand, segmenting volumes, toggling between half a dozen unconnected systems, and dictating reports into software that fought back. By some estimates the administrative drag eats most of a radiologist's shift. The diagnosis - the part requiring a decade of training - takes minutes.
Meanwhile the math is getting worse. The United States is projected to be short somewhere between 10,000 and 35,000 radiologists by 2034. The studies keep coming. The people to read them do not.
So the obvious play - build an AI that replaces the radiologist - was both the most hyped and the most wrong. New Lantern bet on the opposite: keep the radiologist, delete the drudgery. Replace the software, not the doctor.
Fig. A — The most expensive part of a radiologist's day is the part a teenager could be trained to resent.
"Countless companies chased the idea that AI should replace radiologists. Shiva figured out a much better approach is to use AI to replace the drudgery of the job."
- Eric Vishria, General Partner, BenchmarkDuring the pandemic, Shiva Suri - then an engineer who had worked at the data-streaming company Confluent - was holed up at his parents' place, sharing a home office with his mother. She is a radiologist. He watched her, day after day, lose hours to software that seemed almost designed to slow her down.
The thought that followed is the kind most people have and then forget. Suri did not forget it.
Built the early team out of people he already trusted - friends and former colleagues - on the theory that when a startup is hard, trust is the thing that does not break. Advice to other founders: ignore investors early and just ship.
"Some startup is going to fix this and make my mom's life easier. Why can't it be me?"
- Shiva Suri, on starting New LanternSuri leaves the data-infrastructure world to rebuild radiology software from the cloud up.
Benchmark leads; GP Eric Vishria joins the board. Total raised passes $23M. The "AI radiology resident" goes public.
An in-house dictation model tuned for radiology - faster transcription, accuracy that adapts to the user.
The platform expands into specialized mammography modes and nuclear-medicine workflows.
Industry press profiles New Lantern as a single workspace built for the AI era - PACS, worklist and reporting unified.
Fig. B — A four-year sprint from "my mom deserves better software" to a Benchmark board seat. Most people just buy their mom flowers.
Legacy radiology runs on a tangle of separate systems - a viewer here, a worklist there, a reporting tool that talks to neither. New Lantern's argument is that the tangle is the problem. It collapses the stack into a single browser-based workspace.
Sub-second image loading with smart precaching and hanging protocols. No installs, no VPNs - it runs in the browser.
Prioritizes studies by STAT status, modality and assignment across multiple sites, with live RVU and turnaround analytics for admins.
Drafts structured reports in seconds, in the radiologist's own language - up to 75% written before dictation begins. Doctor reviews and signs.
An in-house speech model built for radiology, marketed at roughly 3x speed and ~99% accuracy that learns the user's patterns.
Fig. C — The report engine is named Curie. If you are going to put a patron saint on medical imaging, you could do worse than the woman who discovered radium.
"When everything is working against you as a startup, having people you trust matters hugely."
- Shiva Suri, on building the team from friendsClaims are cheap in healthtech. Here is what New Lantern puts on the record: it says the platform already automates about a quarter of radiology workflows, drafts the bulk of reports before a doctor speaks, and translates that into measurable productivity - the kind administrators count in RVUs.
The backers add a different kind of proof. The $19M Series A was led by Benchmark, with checks from Afore Capital, Anthology Fund, Neo, SV Angel and a roster of operator-investors: Guillermo Rauch (Vercel), Amjad Masad (Replit), Saji Wickramasekara (Benchling), Gokul Rajaram, and Jay Kreps - the Confluent co-founder who was once Suri's boss.
Fig. D — When your former employer writes a check to your competitor for his attention, that is one kind of endorsement. New Lantern got the other kind.
The mission is narrow on purpose: automate the tedious parts so radiologists can spend their time reading. Stated more ambitiously, it is about leverage - letting one radiologist produce something closer to the output of three, which is roughly the gap the country is staring at as demand outpaces the supply of trained eyes.
Notice what the mission is not. It is not "AI will read your scan." A human still signs every report. The FDA registration reflects this - New Lantern is a Class I medical image communications device, the plumbing and the assistant, not the diagnostician. The radiologist stays in the chair. The software just clears the desk.
"One radiologist, three radiologists' worth of output. That is the bet - workflow, not wizardry."
- The New Lantern thesis, in a sentenceReturn to the reading room. The radiologist opened a single tab, read an image, signed a report, and moved on. Multiply that across a shift and the shortage of tens of thousands of radiologists starts to look less like a wall and more like a problem of leverage.
That is the whole story New Lantern is trying to tell: the future of radiology is not a machine that replaces the doctor. It is a machine that gives the doctor back the day. Whether the company hits 90% automation or stalls at 40, the direction is the interesting part - and for once, the radiologists are rooting for the AI.
It started because one engineer could not stand watching his mother fight her software. It is, in the end, a very long way to go to fix your mom's computer. It also might be the right way.