213: Quantitative Regression of qFibrosis with Resmetirom in MAESTRO-NASH Trial

Paper Discussed in this Episode:

Quantitative regression of qFibrosis with resmetirom: Exploratory histologic endpoints from the MAESTRO-NASH phase III clinical trial. Schattenberg JM, Bedossa P, Guy CD, et al. Journal of Hepatology 2026; https://doi.org/10.1016/j.jhep.2026.03.021.

Episode Summary: In this deep dive, we explore how artificial intelligence is revolutionizing the way we measure liver disease recovery. We examine a groundbreaking 2026 Phase III clinical trial (MAESTRO-NASH) that compared traditional human pathologist staging against an AI-driven digital pathology tool called qFibrosis. The study forces us to reconsider our clinical gold standards by asking: what if AI can detect subtle biological healing that the experienced human eye completely misses?

In This Episode, We Cover:

• The Silent Epidemic: Understanding Metabolic dysfunction-associated steatohepatitis (MASH), a progressive, active form of fatty liver disease linked to cardiovascular risk and cirrhosis. We discuss why precisely tracking the reversal of liver fibrosis is crucial for patient outcomes.

• The "Ordinal" Problem: Why the current "gold standard"—human pathologists assigning a simple ordinal score (like Stage F1, F2, or F3)—is subjective and fails to capture the dynamic, nuanced reality of fibrosis progression and regression.

• The AI Microscope (SHG & qFibrosis): ◦ SHG (Second Harmonic Generation): An imaging technique that takes advantage of the physical properties of collagen to map out the three-dimensional architecture of the liver. ◦ qFibrosis: An AI-driven analysis tool that evaluates up to 184 distinct features of liver collagen (like string length, width, and intersections) across different regions of the liver lobule, providing a continuous, hyper-detailed assessment rather than a basic category.

• The Showdown - Humans vs. AI: Using data from 966 patients in the MAESTRO-NASH trial, we compare how human pathologists and the AI evaluated liver biopsies at baseline and week 52 to test the efficacy of the drug resmetirom.

• The AI's "Aha!" Moment (Seeing the Invisible): The most shocking finding of the study occurred in the "non-responder" group. Even when human consensus reads declared certain patients had no histological improvement, the AI detected significant, continuous reductions in liver fibrosis (qFC scores). The digital pathology tool was able to pick up on subthreshold, early matrix remodeling that was entirely invisible to standard manual scoring.

• Mapping the Liver's Healing: The AI proved its biological accuracy by successfully linking its spatial data to real-world clinical outcomes. The AI found that specific regional changes—particularly in the portal tract—strongly correlated with non-invasive liver stiffness tests like Magnetic Resonance Elastography (MRE).

Key Takeaway: AI isn't here to replace human pathologists; it is a hyper-sensitive tool designed to uncover hidden data patterns. By detecting continuous, region-specific changes in liver collagen, AI digital pathology can identify early therapeutic responses to MASH treatments that traditional staging misses, fundamentally changing how we track disease reversal and personalize medicine

Get the "Digital Pathology 101" FREE E-book and join us!

Transcript

Welcome to the digital pathology podcast.

Glad to be here.

To all the trailblazers listening today, uh the health care professionals, the medical pioneers, and you know, the researchers working right at the cutting edge of medicine, we are doing a deep dive into the source material today that well, it fundamentally challenges how we define whether a drug actually works.

It really does. It's a massive shift,

right? So, our mission today is to explore this brand new, genuinely game-changing paper from the Journal of Hepatology. And we definitely need to credit the authors right out of the gate. I mean, we're talking about Uranim Shatenberg, Pierre Bossa,

Cynthia Dye, Rebecca Ta,

right? And of course, the late Steven A. Harrison, his dedication to this field was just unparalleled alongside a really brilliant team of colleagues.

Absolutely. And their work focuses on the Master Nash phase 3 clinical trial. But specifically, it's looking at this piece of technology that honestly, it sounds almost like science fiction to me.

Yeah, it does sound a bit out there. They use artificial int igence and a digital pathology system called fibrosis. They used it to measure the actual microscopic impact of a drug called resmatyum on patients who are suffering from mah.

And look, the reason this specific paper is making such massive waves right now is that it proves this AI system can physically see healing in the human liver that the human eye completely like entirely misses.

Exactly. It detects structural disease reversal in patients who under normal clinical trial rules would basically be labeled as treatment failures

which is wild. So to understand how an AI can overrule a human pathologist I think we have to start by looking at the battlefield itself. We're dealing with metabolic dysfunction associated stoepitis

or uh mah which is just an enormous silent health crisis. I mean in the United States alone we are looking at roughly 6.65 million people affected by this.

Wow. 6 and a2 million.

Yeah. It's a progressive disease and it's It's driven by metabolic dysfunction that leads to chronic liver inflammation.

And the liver doesn't really like being inflamed constantly.

No, it doesn't. In response to that relentless inflammation, the liver tries to protect itself. It does that by laying down scar tissue. That scar tissue is made of collagen. And well, the buildup of that collagen is what we call fibrosis.

Right. The classic fibrosis.

Exactly. And over time, that fibrosis stiffens the organ. It leads to severe outcomes, cardiovascular death, cerosis, and ultimately endstage liver failure. So the entire goal of new therapeutics like resmatum is to halt that collagen production or ideally reverse it.

But developing a drug to melt away scar tissue. Um it brings up a massive logistical hurdle because the hardest part of a clinical trial for MSH isn't just treating the disease. It's actually proving to the regulatory bodies that the treatment worked.

Right? Proving it is the bottleneck because the current gold standard for doing that relies on well a human being looking through a microscope.

The nays h It's a scoring system.

That's the one. And historically, look, this system was incredibly valuable. It gave the medical community a shared language. A pathologist takes a needle biopsy of the liver, stains it with these chemical dyes, so the collagen turns blue, and then they look at the slide,

just visually inspecting it.

Yeah. And based on what they see, they assign the tissue in ordinal stage. You start at F0, which is normal. Then you move up through these discrete stages like F1B, F2, F3, all the way to F4, which is full-blown cerosis.

Okay. So step-by-step ladder.

Exactly. And to get a Maya drug approved, a pharmaceutical company usually has to prove that their drug drops a patient's fibrosis score by at least one full stage without the disease getting worse.

Okay, let's unpack this because using the traditional Nesh HCRN system to measure a drug's efficacy seems incredibly frustrating. I mean, it's kind of like trying to track a patient's weight loss journey using only the holes on their belt.

Oh, that's a perfect analogy. Yes.

Right. Like if a patient changes their diet, they exercise, every day and they lose 5 lbs, well, their blood pressure might drop. Their cardiovascular risk decreases. Their body is objectively healing.

They're getting healthier.

But if they haven't lost enough weight to tighten that belt to the very next hole, the belt still fits exactly the same. So under the rules of the belt, zero progress was made. The measuring tool is just too blunt to capture the nuanced, you know, biological reality of the healing.

That analogy gets to the very heart of the problem with ordinal staging in clinical trials. Liver fibrosis is highly dynamic. It's not a static piece of wood. It's living tissue with this constant turnover of collagen happening at the microscopic level.

Right. It's active.

Very active. Yeah.

But the human eye looking at a chemically stained slide is forced to compress all that complex continuous biological activity into four or five rigid buckets

and adult holes.

Exactly. And add to that the issue of interrator variability. You could have two worldclass experts look at the exact same blue stain slide and disagree on whether it's an F2 or an F3.

Oh man. So, it's subjective on top of being blunt.

Very much so. You even have interrator variability. The same expert might score a slide differently on a Tuesday than they did on a Thursday.

That's that's a huge problem for a billion-dollar drug trial. So, if we want to truly understand if a drug is remodeling the liver, we need a tool that measures the actual pounds of collagen, right? The continuous granular data rather than just waiting to see if we can reach the next belt hole.

Precisely. And that brings us to the methodology. Shatenberg and his team deployed in this substudy of the Maestro Nash trial. They looked at paired biopsies, one taken before treatment and one taken 52 weeks later from 966 patients.

But they bypassed the subjective human eye entirely. They use lasers.

Yeah, they did. They employed a technique called second harmonic generation or SHG alongside two photon excitation fluoresence microscopy.

Okay, I have to pause here. Standard light microscopy requires you to pour chemicals onto a slide to dye the tissue. Right, so you can see But this SHG relies on pure physics.

Pure physics. It's a massive leap from standard pathology because it fundamentally changes how we interact with the tissue.

I want to make sure we ground that concept for the trailblazers listening because terms like second harmonic generation can sound uh pretty intimidating. How does shooting a laser at a biopsy let us see the actual weight of the fibrosis?

Well, it comes down to the physical geometry of the collagen itself. Furular collagen has a very specific nonentroymmetric structure,

meaning

it just means the molecules lack a center of symmetry. And because of that unique physical quirk, when you fire a specific type of infrared laser at the tissue, two photons from the laser hit the collagen molecule simultaneously and combine.

Wait, they combine?

Yeah, they bounce back as a single photon with exactly half the wavelengths and twice the energy. The collagen literally generates its own light signal. It glows brightly against a dark background completely naturally.

Oh wow. Meaning you don't need any chemical stains that might be applied inconsistently and you don't need a human to interpret a specific shade of blue. The laser just forces the physical scar tissue to announce its exact location

and a computer records it perfectly.

That is incredible.

It is. And that data is then fed into the AI system known as gibrosis.

The AI maps the exact three-dimensional architecture of the collagen across the entire biopsy. But, and this is key, it doesn't just measure the total surface area of the globe,

right? It gets hyper granular.

Exactly. It extracts 184 distinct mathematical features from six specific anatomical regions of the liver lobule.

Okay. Those six regions being the portal tract, the periportal region, the central vein, the perententral region, zone 2, and my personal favorite, the incredibly descriptive chicken wire fibrosis.

Chicken wire is a great visual for it. Yeah.

So, within every single one of those zones, the AI is counting the literal number of collagen strings. It's calculating the exact length of those strings, their width, and how many times they intersect with each other to form webs.

It's a staggering amount of geometric data. It basically creates a comprehensive topological map of the disease state,

which naturally leads me to a bit of skepticism. I mean, whenever I hear about an AI extracting 184 different geometric variables from a single image, my immediate thought is data dredging. Like, are we just counting irrelevant pixels here? Does the exact length of a collagen string inter secting in zone 2 actually mean anything biologically or is this just digital noise masquerading as precision?

That is the most critical question you can ask when evaluating any AI diagnostic tool. If the variables don't tie back to human health, the tool is useless, right?

But what's fascinating here is and the authors address this directly. They anchored fibrosis to a foundational piece of previous research called the steight study.

Okay. Stetoy.

Yeah. The developers of this technology didn't just guess which features mattered in ste They took archived historical liver biopsies from thousands of patients, ran this exact AI analysis on them, and then looked at those patients actual medical records 5 or 10 years down the line.

Oh, I see. So, they tracked the microscopic math against realworld survival rates.

Exactly. They mapped those specific features against clinical outcomes like apatic decompensation,

which is when the liver stops functioning and fluid starts building up in the abdomen. Right.

Yes. Very severe. They track that and all cause mortality and they prove definitively that 30 of these highly specific kibrosis features were validated predictors of those severe outcomes.

Wow. So when the AI is flagging say aggregated fiber length in the portal tract, it's not just observing a random pattern. It is quantifying the literal architectural changes that historically we know will lead to a patient getting sicker or dying.

Exactly. That completely validates the tool. It's mapping the architecture of liver failure.

Okay. So we take this validated hypersensitive AI and and we unleash it on the Maestro Nash trial to see if the drug risarum actually works. Let's dig into the clinical results after 52 weeks of treatment.

Right? So the topline efficacy data basically confirmed what the human pathologists were seeing but with absolute mathematical precision.

Patients were given either an 80 mgram dose of resetum, a 100 millgram dose or a placebo.

And after a year

the AI calculated a firosis stage which is its own algorithmic version of the staging system and the imp improvements were massive. Roughly 24% and 22% more patients in the drug groups improved by at least one full stage compared to the placebo group.

So the drug definitively halted the worsening of fibrosis and actively reversed it for a large portion of the cohort.

Yes, the macrolevel success was undeniable.

But and here's where it gets really interesting. The part of this paper that truly reshapes how we think about clinical trials is buried in the continuous score data, what they call the QFC. The exact poundage of the fibrosis. To go back to our earlier analogy,

the researchers isolated a specific group of patients for this, the human non-responders.

These are the patients where the worldclass human pathologists looked at the week 52 biopsy through their microscopes and declared that no meaningful change had occurred.

Right. Visually, the patient had not dropped a full fibrosis stage.

Exactly. The started as an F3 and a year later, the human eye still saw an F3. Under conventional regulatory metrics, that patient is a treatment failure. Yet, when the team looked at the AI's continuous score for those exact same non-responding patients, the narrative completely flipped. The AI detected a statistically significant reduction in continuous fibrosis. We're looking at a negative.8 and a negative.9 reduction in the QFC score for the 80 milligram and 100 milligram drug groups.

And the placebo group for those non-responders, a completely flat 0.0.

Wow. So, the placebo group showed zero continuous improvement while the treated patients showed distinct measurable dissolution of collagen.

So, what does this all mean for you, the trailblazers listening?

Well, if the AI is detecting a nearly onepoint reduction in the continuous score, but the pathologist sees zero stage change, that implies the drug is actively remodeling the cellular matrix and dissolving scar tissue, but it's happening at a resolution the human eye just physically cannot perceive.

Which means we are quite literally denying patients credit for real biological healing simply because they haven't crossed an arbitrary visual threshold.

That's huge.

It is digital pathology detects sub threshold early matrix remodeling. That is the core revelation of the study. It proves the anti- firootic efficacy of resmatum on a microructural level. The ordinal staging system is just blind to this continuous biological healing. The drug is working. The tissue is recovering. But our manual measuring stick was too blunt to capture it.

And the fibrosis system doesn't just tell us that the liver is healing, right? It tells us exactly where the healing originated.

The geography of the remodeling. Yes. And And that geography provides massive clues into how the drug actually operates because the AI showed that Resmaterum doesn't just uniformly melt collagen across the entire organ all at once.

No, it's very localized. Initially, the strongest, most significant reductions in those dangerous collagen features occurred specifically in the portal tract, the periportal regions and within those chicken wire fibrosis networks.

Understanding why the portal tract clears first seems crucial here.

It is. The portal tract is a essentially the main entry point for blood coming into the liver's lobules. It houses the portal vein, the hpatic artery, and the bile duct.

Okay, let's visualize that. If we imagine the liver as a massive manufacturing complex, the portal tract is like the primary sixlane highway supplying the factories.

That's a great way to picture it.

So, in a diseased state like mah, those factories are producing toxic amounts of scar tissue, causing massive traffic jams of collagen all along those main supply routes. If a drug successfully shuts down the run scar tissue production inside the factory. It makes total sense that the supply trucks clear off the main highway, the portal tract first, long before the tiny side streets and back alleys deeper in the tissue begin to clear out.

That mechanism based analogy translates perfectly to the clinical data. And look, the reason we care so deeply about the clearance of that specific highway is because of how it correlates with the patients actual physical health on a macroscopic level.

How so?

Well, the researchers took these specific regional AI fe es like the shrinking aggregated fiber length in the portal tract and they correlated them with non-invasive clinical tests or NITs.

Right? These are the tests hepatologists use in standard practice to check on a liver without having to physically puncture the patient with a biopsy needle. We're talking about advanced imaging techniques like MRE, which is magnetic resonance elastography, and VCTE, commonly known as fiber scan.

Exactly. And both of those technologies measure the physical stiffness of the entire liver. MRE, for example, uses these specialized acoustic waves that vibrate through the liver while the patient is inside an MRI machine.

Sounds intense.

It creates a visual map of tissue elasticity. And the rule is simple.

The stiffer the liver, the worse the fibrosis.

Okay.

The study found incredibly strong positive correlations between the AI's microscopic portal tract data

and the MRI's macroscopic stiffness data.

So, as the AI watches the microscopic collagen strings dissolve on the main highway, the RI machine is simultaneously feeling the entire 3 lb organ physically softening up.

Exactly. The micro directly explains the macro.

To fully appreciate why this specific drug is capable of clearing that highway though, I think we need to dive into the mechanism of resarum itself.

Yeah, let's do it. Resmatarum is a liver directed THR beta agonist.

THR beta.

THR beta stands for thyroid hormone receptor beta. In a healthy human liver, the THRB regulon plays a massive role in managing metabolism, particularly how the liver handles lipids and fats.

Okay. So it's a fat manager,

right? But as a patient develops MASH, the liver becomes completely overwhelmed by metabolic stress and lipotoxicity.

The liver is literally under attack from fat. So its natural defense systems start shutting down or malfunctioning.

Exactly. And in response to that overwhelming toxic stress, the body actually downregulates or decreases the activity of that normal THRB metabolic pathway.

It just turns it down.

Yeah. And when that THRB activity drops, it removes a critical regulatory check on the liver's cells. It unleashes these fibrogenic genes, specifically genes known as AEBB2 and BNC2.

So you can think of these genes as like runaway foreman at a construction site without the THRB pathway acting as the site manager to keep them in line. These foremen panic and start ordering the mass production of collagen. They just lay down endless amounts of scar tissue regardless of whether the tissue actually needs it.

That's exactly what happens. So by introducing resmatyone which actively agonizes or stimulates the THR beta receptors, we are essentially parachuting a new site manager into the liver.

The drug artificially ramps the THRB pathway back up to healthy levels,

right? And that revitalized pathway basically fires the runaway foreman. It directly suppresses the AEBP2 and BNC2 genes. It puts the brakes on the fibergenic pathways and stops the runaway collagen production right at the genetic level.

And the absolute genius of this entire Master Nash sub study is that Shatenberg and his colleagues proved We now have the technology to watch that exact genetic mechanism play out in real time. I mean, we are tracking a pill that tweaks a specific thyroid receptor. We are visualizing its effect on microscopic collagen strings in exact zones of the liver using non-chemical lasers.

And we are proving that those invisible microchanges perfectly correlate with the physical softening of the organ and the ultimate survival of the patient. It represents a total triumph of modern pathology.

It really does.

This paper unequivocally demonstrates that continuous fibrosis metrics are not just some optional upgrade. They're absolutely essential if we want to accurately map the pathophysiology of disease progression and regression in metabolic liver disease.

So to bring this all home for the trailblazers tuning in today by applying AI and second hormonic generation lasers to thousands of biopsies. This team didn't just prove that resmatarum is an effective treatment for MSH. They completely redefined the resolution at which we are capable of viewing human healing.

They dragged the field from a clunky manual subjective staging system into an era of precise continuous biological metrics.

Digital pathology is clearly no longer just a neat tech demo discussed in conference halls.

No, not at all. It is actively revealing subtle life-saving treatment effects that our conventional methods are just entirely blind to.

Which leaves us with a final lingering thought to mull over today as we look toward the future of clinical trials. You know, if digital continuous metrics can definitively detect structural healing long before a human pthologist can perceive a shift in an ordinal stage. How soon until regulatory bodies abandon the old staging system altogether?

It's a matter of time,

right? And when they finally make that leap, how many failed drugs of the past might actually turn out to be incredibly effective if we just looked at them through the right lens? It's something to think about. Keep pushing the boundaries, keep asking the hard questions, and most importantly, stay curious. Thanks for listening to our deep dive today.