A urologist clicks open a digital slide. Pinks, purples, the chaotic geometry of disease. Somewhere in that picture is the answer to a single, brutal question: will this man die of this cancer, or with it? For most of medicine's history, the answer was a shrug dressed up in statistics. ArteraAI is trying to make the shrug obsolete.
The quiet startup that moved the guidelines
ArteraAI is a precision medicine company that builds multimodal AI tests for cancer. That sentence does a lot of work, so let it. Multimodal means the algorithm reads both the digital pathology image and the clinical chart at the same time, the way a thoughtful oncologist would, if she had infinite hours and a photographic memory. Precision medicine means the goal is therapy tuned to one patient, not a cohort. And cancer, well, cancer is the part that makes the rest matter.
Today the company occupies a narrow but extraordinary perch. Its flagship product, the ArteraAI Prostate Test, is the first AI-enabled predictive and prognostic test ever included in the NCCN Clinical Practice Guidelines for prostate cancer - the national rulebook that oncologists actually consult. It earned the highest tier of supporting evidence the guidelines offer. It carries an FDA De Novo authorization and Breakthrough Device Designation. It runs from a slide that already exists in a hospital archive.
Cancer care, written in crayon
Prostate cancer is the second most common cancer in American men, and one of the more confounding. Most men diagnosed with localized disease will not die from it. Many will. Distinguishing the two groups in advance has been, to put it gently, an art. Risk stratification uses PSA blood tests, Gleason scores from a pathologist's eye, and tumor stage - tools that were largely invented before the iPhone.
The result: a lot of men get treated who didn't need it, and a smaller, sadder group gets watched when they should have been treated. Radiation, surgery, hormone therapy - none of these are free. They cost continence, potency, time. Picking the wrong one is the kind of mistake a patient lives with for a decade.
The team behind Artera believed the missing signal was hiding in plain sight. It was sitting on the glass slide that every prostate-cancer patient already has. Pathologists were reading those slides with their eyes. A neural network, properly trained, could see things eyes could not - patterns of stroma, gland shape, micro-architecture - and turn them into a number a clinician could act on.
The bet, in one line
The same biopsy slide already in your medical record contains the signal. You don't need to extract DNA, draw new blood, or wait two weeks. You need a better reader.
Andre Esteva, encore
If the name Andre Esteva rings a faint bell, it is because in 2017 he was the lead author of a Nature paper that startled the medical world: a convolutional neural network classified skin cancer at the level of board-certified dermatologists. He was a Stanford PhD student at the time. He spent the next chapter at Salesforce Research running medical AI. Then he left to build Artera.
His co-founder, Dr. Felix Y. Feng, is a radiation oncologist at UCSF and one of the more cited researchers in prostate-cancer biology. The combination matters. Esteva brings the model architecture; Feng brings the trials, the slides, and the credibility with the panels that decide what becomes the standard of care. Without him, the company would have been another AI demo. With him, it became evidence.
Five years, one short list
One slide. Two models. A decision.
The ArteraAI Prostate Test is not one model, it is two doing different jobs on the same image. A prognostic model estimates long-term outcomes - distant metastasis, prostate-cancer-specific mortality, overall survival. A predictive model estimates whether a specific therapy, in this case hormone therapy added to radiation, is likely to help this particular patient. Predictive and prognostic are different verbs, and oncologists have been waiting for a tool that does both.
Prostate Test (Biopsy)
The flagship. Runs from a digitized core-needle biopsy. Outputs a risk group and a predicted therapy benefit. FDA De Novo cleared.
Prostate Test (Post-RP)
For patients after radical prostatectomy. Estimates risk of biochemical recurrence to guide post-surgery treatment.
Active Surveillance Module
Helps clinicians decide which men with low-risk disease are safe to monitor, and which are not.
Why the NCCN moved
Receipts, in five Phase 3 trials
Most AI in healthcare lives in a slide deck. ArteraAI lives in a slide. The distinction is funded, peer-reviewed, and frankly inconvenient for competitors. The model was trained and validated against the kind of dataset that medicine treats as sacred - the long-followed, randomized cohorts of NRG Oncology. Five separate Phase 3 trials, decades of follow-up, thousands of patients.
The training set is also notable for what it doesn't have: a white-only blind spot. ArteraAI has made a point of validating the model on race-diverse cohorts, because the failure mode of medical AI is, almost always, that it works splendidly on the population it saw and embarrassingly on everyone else. That choice will age well.
Cancer is personal. The therapy should be too.
The company's own line is also its mission, and unusually for a tagline, it survives close reading. The unifying argument is that cancer treatment got stuck in a one-size-fits-most rut because the tools used to stratify patients were too coarse. Fix the tools - with AI that can see what humans cannot - and you can give every patient the closest thing to a custom plan.
That is the throughline from Esteva's dermatology paper in 2017 to the prostate model today. The patient sitting in a urologist's office in Topeka or Tallahassee shouldn't get worse decision support than the patient sitting in a referral center in Boston. The slide is the same. The algorithm is the same. The standard of care, for once, can be the same too.
Back to the slide
Return to the opening scene. The urologist clicks open the digital pathology image. The pinks and purples are still pinks and purples. But now a second pane is open beside it - a model running in the background that has read tens of thousands of similar slides paired with two-decade outcomes. Within minutes she has a risk score and a predicted therapy benefit she can show the patient and his wife, who came to the appointment terrified and will leave with a plan that fits them specifically.
The shrug is still there. Medicine never gets rid of uncertainty - it just gets better at sizing it. ArteraAI's quietly subversive contribution is that the size now comes from a machine that has done more reading than any single doctor ever could. And the next cancer is already in the pipeline. Breast is the obvious one. After that, the company has the platform problem of every horizontally useful tool: which slide do we read next?
Most healthcare AI promises a revolution and ships a chatbot. ArteraAI shipped a guideline change.