How did a chance conversation in a molecular pathology lab lead to the development of an AI-powered molecular oncology platform? And how might it open up access to precision medicine?
Here, Travis Wold, CEO of Imagenomix, shares how AI is being used to unlock the molecular signals hidden within standard H&E slides, and what it could mean for clinical and biopharma settings.
In the current molecular pathology workflow, what are the factors responsible for “tissue failure”?
Deep tissue biopsies need to be minimally invasive, which means very fine needles and very small tissue samples — often just enough for diagnosis. From there, the sample is divided across multiple stains and diagnostic tests, and in many cases the tissue is exhausted before it ever reaches molecular characterization by next-generation sequencing (NGS).
In roughly a quarter of oncology cases in the US, the diagnostic workflow consumes all of the available biopsy tissue. That’s 25 percent of cancer patients who miss out on molecular characterization entirely — and that’s in the country with the highest molecular testing volume in the world.
The picture gets worse when you zoom out. In the US, an estimated 75 percent of patients still don’t receive NGS-based mutational testing, largely due to cost, turnaround time, and access. Outside the US, that figure climbs to roughly 98 percent — driven almost entirely by cost. Our goal is to extend access to every patient, regardless of income or geography, using nothing more than a standard H&E slide.
How might the AI-driven molecular oncology platform increase access to molecular testing?
Initially, we see our platform, Imagenomix Predict, being used as a screening tool. It assesses the probability of an actionable mutation and flags only the high-probability cases for confirmation with a PCR or NGS gold-standard test. Identifying a targeted mutation early can be life-changing — and in many cases, life-extending.
The cost implications are significant. A single molecular test can run $4,000 to $6,000. An AI prescreen can rule mutations in or out before that spend is committed, accelerating time to treatment and lowering cost per patient. With the right partnerships, our platform could get patients onto targeted therapy weeks faster than the current model — and in oncology, weeks matter. Better answers, delivered faster, lead to better outcomes.
How might the technology be used to improve clinical trial efficiency?
Consider a trial that needs to screen 1,000 candidates against an acceptance criterion of a 5 percent mutation rate. Under the current model, you might spend $5 million on NGS tests just to identify your 50 eligible patients — before the trial even starts.
With AI prescreening, you could rule out 90 percent of candidates upfront and only pay for confirmatory NGS or PCR testing on the remaining 100. And for the patients earmarked for molecular testing at the start of the diagnostic process, you can preserve their tissue. Under the current workflow, histology testing runs first, which is why 25 percent of potential trial candidates are lost to tissue failure before molecular testing even begins. Multiply that against a 5 percent prevalence rate and trial enrollment becomes genuinely difficult.
Speed matters as much as cost. Our platform can run a mutation scan in about three minutes on the same day the H&E slide is prepared. An NGS test takes 7 to 14 days in-house, or 30-plus days through centralized labs — by which point many patients have already started therapy and are no longer eligible for targeted-therapy trials.
A meaningful share of oncology trials fail or stall due to some combination of tissue insufficiency, cost, and turnaround time. That’s the bottleneck we’re working to address as we advance the accuracy and validation of AI-based cancer testing.
What inspired the Imagenomix AI model development?
It started with a conversation in the lab at NYU Langone. The team was talking through future projects when Matija Snuderl, a molecular pathologist, looked at a slide and said, “I think this cancer has an EGFR mutation.” Aristotelis Tsirigos asked him how he knew. Matija said, “I don’t know — I’ve just seen so many of these cases, watched the molecular results come back, and looked at the slides again. I think I can recognize the patterns.”
Aris’s reaction was: if a pathologist can pick out oncogenic driver mutations by eye, AI should be able to do it even better. The challenge was on.
How was the concept developed?
The team built, trained, and tested an algorithm that takes a digital image of an H&E slide and applies machine learning for pattern recognition, using NGS gold-standard results as ground truth. It worked — identifying EGFR, KRAS, STK11, and other common lung cancer mutations directly from histology.
The foundational work was published in Nature Medicine in 2018. By 2020, we had patents on both the algorithm and the methods.
How are you approaching the translation of concept to product?
The medical community isn’t yet at the point where reimbursement codes — or full clinical trust — exist for AI in final diagnosis. So our first focus is clinical trials, where the efficiency case is immediate and measurable.
Biobanks are another strong fit. They hold tens of thousands of archived tissue samples without molecular annotation. They can either spend millions running NGS across the entire library, or use a platform like ours to pre-screen and reduce the NGS run by 80 to 95 percent.
We also see a real opportunity to partner with image management systems like Proscia and PathPresenter. They’re excellent at hosting and managing scanned slide data — but in some ways those systems are like an iPhone with no apps. Our platform layers in as an add-on application that makes their data more actionable. Image management platforms are already integrated into thousands of hospitals globally; we bring the mutational and cancer stratification layer to enhance patient care within their existing workflows.
As AI becomes more embedded in pathology — as trust grows, studies validate, and reimbursement pathways open — we’ll expand into the clinical diagnostic setting. Long-term, we want the platform to serve as a companion diagnostic.
We recognize the current limitations. The progression will be from trust in single-mutation prediction, to small panels, to large panels. AI has advanced enormously in the last two years, and our accuracy with the patented platform is improving in step.
What have preliminary studies shown about the system’s performance?
We’re actively advancing two assays on the patented platform. Our lung cancer assay predicts common driver mutations, forming the foundation of our patent portfolio. The second targets common mutations in breast cancer. Both are on track to launch in Q3/Q4 of 2026.
We also trialed our glioma assay on a specific brain cancer mutation that has an approved targeted therapy but no approved companion diagnostic. In our internal validation, the platform achieved 95 percent accuracy in identifying the mutation. If that result holds through external validation, it represents a meaningful pathway to targeted therapy for patients who are currently being missed.
What’s your vision for AI-powered precision oncology?
Our mission is to make precision oncology available to every patient, regardless of income or insurance status. Targeted therapies save lives, and they are dramatically underused because the testing infrastructure to identify candidates simply isn’t accessible enough.
And we don’t just mean access in affluent markets like the US and Europe — that’s the easier part of the problem. We want to bring mutational cancer testing to the entire world.
It sounds ambitious, but I think we can drive the cost low enough to make it viable in low- and middle-income countries. There’s no logic in launching in the US at thousands of dollars per AI test and then negotiating down country by country. We can start in the hundreds from day one.
Importantly, I don’t see AI replacing pathologists — I see it supporting them. With a shrinking global pathologist workforce, the profession has to work smarter, and platforms like ours are how that happens. Within five years, our goal is to give millions of patients per year access to targeted cancer therapy via AI screening.
Newsletters
Receive the latest pathologist news, personalities, education, and career development – weekly to your inbox.
