Paige

AI Cancer Detection

Andrew Kuzemczak
April 22, 2025

Hi friend,

Welcome back to Future Human! I apologize for taking last week off while I was away. We are now back, I am rested (sort of), and most importantly, I am ready to tackle the remaining eight weeks of M1 year.

We are left with just hematology, endocrine, and reproduction. Then comes summer, where I will just keep working on research, Future Human, and my knowledge of Manhattan’s best running routes. Gone are the summer days with nothing to do but argue with my mom over wearing sunscreen and crush a dozen freezer pops. With that said, I am still very excited for my first summer in NYC.

Okay, back to Future Human.

For the big #10, I wanted to look back at AI. Really, I wanted to investigate a trend that everyone believes will flip medicine on its head in the coming 5-10 years. When I sit down with most non-experts and discuss the intersection of medicine and AI, most will label radiology or general administration as the leading replaceable role in modern medicine. When I chat with my professors, however, they take a more nuanced view.

Inspired by the takes of my professors, I thought I would dive a little deeper into how AI is changing medicine. In my industry-wide search, I was struck by a new target — pathologists.

For those who do not know, pathologists study fluids, tissues, or organs taken from the body. Think, when a biopsy is taken, these are the physicians that review the tissue and tell you what is wrong. They review countless slides from around the body to help us better understand infections, inflammation, and cancers.

With that, let me ask you:

If an algorithm could deliver a cancer diagnosis faster than traditional methods, would you trust its verdict—or would you seek a second opinion from a living and breathing doctor?

The Story

If diagnostic radiologists are adopting AI to better assess radiological images, surely there is room for pathologists evaluating digital slides to implement the technology. Sadly, I was not the first person to think like this.

I first heard of Paige AI in June of 2024 when I read an article on Dr. Thomas Fuchs, then the Dean and Chair for AI and Human Health at Mount Sinai and professor for AI at the Icahn School of Medicine. Although they waitlisted me, I will speak objectively on their former professor, do not fret.

While working at the school in 2017, Dr. Fuchs founded Paige AI with colleagues uptown at Memorial Sloan Kettering Cancer Center (David Klimstra, Norman Selby, and Peter Schüffler). At the intersection of cancer and AI, they thought surely there was something they could do to accelerate either diagnosis, treatment, or both.

Fast forward 8 years, and Paige has become the leader in digital pathology, securing the first FDA approval and breakthrough designation for its computational pathology tech. Paige’s diagnostic AI serves as a co-pilot for pathologists, reviewing tissue samples and alleviating time pressure by automating many tasks in the cancer diagnostic process. The Paige platform runs through the biopsy slides first, delivering insights to pathologists and labeling high priority patients to assist in stratification for quicker and more accurate diagnosis.¹^,²

Not only does the tech bring cancer diagnoses to patients sooner, it also makes each pathologist more confident in their diagnosis. More accurate decisions, delivered sooner — a very promising combination.

A few months ago, I wrote on LinkedIn that Dr. Fuchs was even chosen to serve as Chief AI Officer at Eli Lilly. Although this may appear as just another title for an already impressive individual (he has held positions at Memorial Sloan Kettering, NASA JPL, and CalTech), I believe it gives us a hint as to where this technology may move long term, but I will leave that for the tech section. For now, just consider that Eli Lilly is a global top 10 pharmaceutical giant in oncology (measured by revenue generated by oncology therapeutics). Why would they care about his expertise in AI-driven pathology tech?

The Tech

Having studied the pathology of each organ system I have covered so far in medical school, I can only hope—for the pathologists’ sake—that this technology reaches them as soon as possible.

Paige’s technology is impressive and has improved rapidly just in the last year that I have followed them. In 2024, their model was trained on 4 million slides from 15,000 patients. Today, the model is trained on 25 million slides from 350,000 patients and 200,000 cases with outcome and genetic data. These leaps have allowed them to expand into new cancer types, offer patient screening functionality, and dive into research/drug discovery alongside academic and private institutions. Let me explain.

Paige AI now offers the below tools for various cancer types:

Paige Prostate Suite - aid in the detection and grading of prostate cancer on H&E-stained whole-slide images of prostate needle biopsies
Paige Breast Suite - support identification and classification of breast cancer from breast biopsy and excision specimens
Paige GI Suite - assist pathologists with benign and malignant conditions across the entire gastrointestinal tract
Paige Pan-Cancer Detect - first of its kind tool that can detect regions of interest suspicious for cancer across multiple organs and different tissues, including GI, GU, lung, cervix, endometrium, breast, skin, brain and rare cancer variants³

These 4 cancer tools are brought together with Alba™, a platform that integrates pathology, radiology, and clinical data into a single, AI-evaluated patient profile. Alba™ can analyze pathology images, label areas of suspicion across cancer types, provide clinical insights, screen for molecular biomarkers with Paige OmniScreen™, and summarize all data for the attending physicians.

Paige Alba™ user interface

Diving deeper, I found the core software products that everything runs on. Introducing Paige PRISM™ and Paige Virchow™.

PRISM™ is the software most pathologists will interact with. Its viewer and case management function is key in the digital pathology workflow. The system overlays the AI insights onto the slides during the review so you (the pathologist) can see the tissue of highest interest rather than spending 2x the time searching across the sea of cells.

Virchow™ is more the research engine and platform for developing new AI tools, built in collaboration with Microsoft. Trained on over 1.5 million slides, the self-supervised transformer model (sounds terrifying) aims to excel in detecting cancer across any tissue type (no more need to input a breast biopsy image into the specific breast AI suite). Virchow™ will simultaneously be able to infer biomarker status simply from tissue images, assisting in personalized therapy decisions.

Long term, and the reason I believe Eli Lilly tapped Dr. Fuchs for their C-suite, this technology will accelerate drug development. Just imagine, a clinical trial where patient cancer screening for inclusion or exclusion can be run automatically. From there, biomarkers will be noted by the AI, highlighting which patient will respond best to each treatment, just from their pathology slide. Old treatments can be repurposed in new patients — at the advice of an algorithm. Shocking, but promising tech.

The Market

As much as I wish I was only the second person to think of applying AI analysis to pathology slides after Dr. Fuchs, that would radically inflate my capabilities. Other scientists have taken action, albeit to a smaller and lesser talked about degree than Paige AI.⁴

Many are trying to streamline pathology in all sorts of ways. To simplify our analysis, I ignored those trying to apply other tech besides AI to slide analysis — think shrinking image file sizes to simplify sharing (Proscia) or building machines to make glass tissue slides automatically (Clarapath).

In the computational pathology sphere, there remain a lot of competitors:

PathAI - founded in 2016
- Raised over $250M; strong pharmaceutical and clinical partnerships
Ibex Medical Analytics - founded in 2016
- Originated in Israel; FDA Breakthrough Designation for prostate AI
Aiforia - founded in 2013
- Originated in Finland; went public on the Helsinki Stock Exchange
Deep Bio - founded in 2015
- Originated in South Korea; prostate cancer grading leader in Asia
Owkin - founded in 2016
- Originated in France; big on multimodal AI and federated learning

Despite the sheer number of competitors, Paige AI reigns supreme based on FDA traction, clinical deployment, partnerships, and data infrastructure. Let’s dive in.

As mentioned earlier, Paige secured the first FDA-approval for an AI pathology product with their prostate cancer tool in 2021. From there, it is worth noting that their model is based on one of — if not the — largest digital slide datasets in the world, developed in collaboration with Memorial Sloan Kettering (MSK). I will admit I am partial to MSK, considering I am looking right at the building as I write this from the Weill Cornell student lounge. Separately, Paige also has remarkable partnerships, including Microsoft on the software side and Eli Lilly and Janssen on the pharmaceutical front. Finally, they are moving the fastest toward a generalizable model. The future, in my opinion, is with the AI tools that can assess any biopsy across tissue and cancer types, not with those that specialize in one organ.

I would venture to say PathAI and Ibex are their closest competitors. PathAI boasts more pharmaceutical partnerships, including GSK and Bristol Myers Squibb. They also made a big push toward diagnostic giants, striking deals with Labcorp and Quest for their pathology workflows. They are also well integrated in the drug development pipelines of their pharmaceutical partners, but they lack the biomarker capabilities of Paige. Ibex, on the other hand, is well established in Europe and Asia. They are more known for their value in routine diagnosis as compared to research and clinical trials.

For Paige, pole position will only be maintained with more advanced drug development tools, international expansion, and key partnerships across the private sector.

Thankfully, in a growing market, there is room for many players. Late last year, PathAI even announced a partnership with Paige, offering a place for the Paige algorithm directly within their AISight Image Management System.⁵ The market is large, growing, and they all know it.

The traditional pathology market is massive—think LabCorp and Quest. The global pathology laboratories market was valued at $386 billion in 2024 and is expected to grow at a CAGR of 8.1% until 2030.⁶ Digital pathology remains comparatively tiny, telling me the opportunity to grow and absorb the traditional market is tremendous. The global digital pathology market was worth $1.11 billion in 2024 and is projected to hit $2.43 billion by 2034.⁷ Just think — old fashion pathology must catch up eventually, and it represents a nearly $400 billion opportunity for AI.

We have a leading company with top AI talent and the largest database, operating in a small market — with the opportunity to disrupt a $400 billion industry. I am staying to watch the end.

The Sick

In 2025, there will be an estimated 2,040,000 new cancer diagnoses in the U.S. (5,600 each day) and 618,120 cancer deaths.⁸^,⁹ For all of these individuals, it does not take much to recognize what Paige’s tech could mean for each of their stories.

With their AI able to pre-screen and flag suspicious slides for pathologists in minutes, Paige can help diagnose patients faster. In the world of aggressive cancers (prostate and breast), this can preserve life. In a study published just last month, Paige showed a 55% average reduction in reading time for pathologists. The algorithm highlighted areas suspicious of harboring metastasis. Three pathologists were asked to review a dataset comprising 167 breast lymph node images, of which 69 harbored cancer metastases of different sizes. In suspicious slides, reading time fell by 69.1% with AI-assistance.

Speed is not the only goal. With the product trained on millions of annotated slides, Paige’s tool reduces human error and delivers more accurate diagnoses. The second set of “eyes” is critical in early-stage and micro-metastatic cancers. All together, we are left with fewer false negatives (missed cancer) and fewer false positives (unnecessary stress or treatment). The study mentioned above revealed a sensitivity jump with the help of AI from 73.3% to 90.0% in micro-metasteses.

With quicker and more accurate diagnoses demonstrated, Paige is now trying to offer more detailed and stratified analyses. For those with any experience in oncology, you know that subtyping and risk stratification is everything.

Lung Cancer TNM Staging System

Paige is rapidly developing tools that do not just say “cancer” or “no cancer”, but rather tell you the type and aggressiveness of it. Treatment varies wildly based on the grading that a pathologist assigns, so any improvement in this process could be life altering for patients and physicians. More accurate cancer staging means less over- or under-treatment, and more appropriate use of surgery, radiation, and surveillance.

Next, AI models do not sleep, burn out, or remain tied to certain geographical locations. For underserved regions with a shortage of pathologists, Paige can allow each physician to evaluate multiple more patients — further advancing patients to the treatment stage to save more lives sooner.

Finally, as with the drug development capabilities mentioned earlier, the future of AI in pathology could see medicines improved for each individual. Paige is moving to combine pathology with genomics to help decide on precision therapies for each patient. One day soon, a cancer patient will receive not just a quicker diagnosis thanks to AI, but also the exact right therapy tailored to their genes and conditions.

The Economy

I have written many times before, funny enough in my cardiovascular newsletters more than anywhere else, how cancer care costs are massive (still second to cardiovascular disease, but big regardless). But it is not hopeless — for all the research we have put into understanding cancer costs, we have also uncovered trends in cost reduction.

According to the WHO, studies in high-income countries have shown that treatment for cancer patients who have been diagnosed early are two to four times less expensive compared to treating people diagnosed with cancer at more advanced stages.¹⁰ How then do we diagnose early — you ask?

The WHO says equipping health services and training workers are the key steps to accelerate cancer diagnosis — admittedly not a jaw-dropping discovery. What if AI could help the few trained workers they already have? Maybe multiply their impact?

“To meet the cancer needs of developing countries, we need around 10,000 additional radiation oncologists, 6,000 medical physicists, and 20,000 radiation therapists”

Yukiya Amano, Director-General of the IAEA,

With that said, some may automatically leap to the feared elimination of those roles by AI. I have heard it time and time again. These AI tools will wipe out pathologists, radiologists, and general practitioners.

It is a reasonable fear, but I am confident this will not be the case for decades to come. What will arrive sooner, however, is the replacement of clinicians who do not understand AI by those who do, so get studying medical students.

I realize now that I am in the economy section and have yet to lob some dollar amounts at you, so let us begin.

In 2023, a massive paper from Dr. Simiao Chen and colleagues projected the economic cost of 29 cancers in 204 countries and territories. According to the analysis, the global macroeconomic cost of cancers was estimated to be $25.2 trillion from 2020 to 2050. This equates to an annual tax of 0.55% on global output, or a per capita burden of $2,857 international dollars (INT$) during the same time frame.¹¹ High-income countries face the greatest economic burden, at 0.72% of total GDP, compared with 0.26% of GDP for low-income countries.

So to restate the obvious, cancer is expensive for everyone. We know early diagnosis saves lives most importantly, but also saves money long term. With radiology and pathology being the leading confirmatory diagnostic methods, any effort to improve and accelerate those services can have substantial impact.

Paige AI is set as the leader in the AI pathology realm, poised to save countless lives and slash healthcare costs globally if well adopted. Best of luck to them!

My Thoughts

In the world of healthtech, I have always claimed to be hardware-focused. In my earlier newsletter #7, I explained how I view devices in, on, or around the body as one of the most direct ways to save a life through innovation. I must say, however, after 10 newsletters and months writing on LinkedIn, I have found a new appreciation for other approaches including LLMs/machine learning, biotech, and artificial intelligence. I cannot wait to keep diving in and adjusting my view on all the ways to save a life.

In a world where cancer lags only slightly behind cardiovascular disease as our leading cause of death—and where billions are invested, and too often wasted, in its diagnosis and treatment—more must be done. Rudolf Virchow, the father of pathology, once said, ‘Medicine is a social science.’ If only he could see his namesake now — helping pathologists and patients one pixel at a time.

To more lives saved,

Andrew

I always appreciate feedback, questions, and conversation. Feel free to reach out on LinkedIn @andrewkuzemczak.