Chat GPT.
###
Here is a detailed, publication-ready article that separates the webinar’s recurring themes from the developments that are genuinely new.
Digital Pathology at a Crossroads—Again. But This Time, Some Things Really Are Different
Digital pathology has been approaching an “inflection point” for so long that the phrase has nearly lost its meaning.
For more than a decade, industry panels have described the same obstacles: too few laboratories are fully digitized; scanners and image-management systems do not always work smoothly together; regulatory pathways remain cumbersome; laboratories struggle to justify the capital investment; and reimbursement has failed to reward either digitization itself or many of the computational tools built on top of it.
The speakers at a recent DeciBio webinar, “Digital Pathology at a Crossroads: Consolidation, Clinical Integration, and What Comes Next,” openly acknowledged this history. Moderator Katie Maloney noted that the field has repeatedly said it is “almost there,” only to leave the decisive breakthrough somewhere in the future.
Yet the panel also argued that the conversation has changed meaningfully over the past one to two years. The question, as one speaker put it, is no longer whether pathology will become digital. It is how quickly—and through what commercial, regulatory and clinical model—the transformation will occur.
That distinction matters. The webinar did not reveal that interoperability, reimbursement and laboratory workflow are suddenly solved. They are not. Rather, it suggested that several developments are converging:
major diagnostics companies are acquiring computational-pathology businesses;
digital adoption is moving beyond elite academic centers;
foundation models are materially changing product development;
computational pathology is moving from workflow assistance toward novel prognostic, predictive and companion-diagnostic uses;
and the first commercially consequential AI-enabled companion diagnostic could create a reason for laboratories to digitize that is stronger than operational efficiency alone.
These are not entirely new ideas. What may be new is that they are beginning to reinforce one another.
The Most Visible Change: Consolidation
The clearest new development is consolidation.
The panel repeatedly returned to Roche’s acquisition of PathAI and Tempus AI’s acquisition of Paige as signals that digital pathology is entering a more mature phase. Smaller computational-pathology companies have produced important innovations, but they have often lacked the global commercial infrastructure, installed hardware base, regulatory resources and customer relationships required to distribute those innovations broadly.
Deirdre DeSaino of Tempus argued that companies such as PathAI and Paige helped create the technology, but that larger partners are needed to carry it into routine use. Nick Brown of PathAI similarly emphasized the combination of advanced technology with Roche’s global reach and laboratory presence.
Consolidation therefore is not simply a financial-market story. It addresses a practical distribution problem. A sophisticated algorithm has little commercial value if the developer must negotiate separately with every scanner company, image-management vendor, laboratory network and national regulatory system.
As Ashley Easter of AstraZeneca explained, small algorithm developers want access to as many customers and software platforms as possible. Large hardware and platform companies, meanwhile, want a sufficiently broad algorithm portfolio to make their scanners and image-management systems more attractive. Consolidation and partnership can serve both sides.
There is also a basic customer-experience argument. Laboratories have faced an exhausting number of scanners, image-management systems and algorithms to evaluate. Health systems may spend substantial time issuing requests for proposals, reviewing technical claims, evaluating integrations and deciding whether a vendor will remain viable. A smaller number of better-capitalized, more comprehensive suppliers could simplify those decisions.
The Consolidation Paradox
The danger is that consolidation could produce closed ecosystems.
A company that controls scanners, staining systems, image-management software and algorithms might have both the incentive and the technical ability to favor its own products. A theoretically integrated “end-to-end” system may become a silo.
The panel nevertheless concluded that the market may already be too fragmented for a truly closed model to succeed. Laboratories have installed equipment from multiple manufacturers, and many institutions will not replace an existing scanner or image-management system merely to gain access to one algorithm. Vendors that restrict compatibility would therefore surrender a large portion of their potential market.
The emerging commercial requirement is consequently paradoxical: companies are becoming larger and more integrated, but their products must remain broadly interoperable.
“Everyone has to be everywhere,” as Easter summarized it.
That may be one of the webinar’s most important conclusions. Consolidation is not eliminating the need for interoperability. It is making credible interoperability more commercially important.
Adoption Is Expanding Beyond the Usual Early Adopters
A second potentially meaningful change is the composition of the adopting laboratories.
Digital pathology initially gained traction at major academic medical centers, where research funding, subspecialty expertise and institutional prestige could justify experimentation. The panel said adoption is now appearing in large reference laboratories, community health systems and independent pathology practices.
That is important because these organizations tend to be more demanding about financial return. They are less likely to digitize simply because the technology is interesting.
Brown described reference laboratories as “hyper focused on financial ROI.” Their willingness to make substantial investments suggests that digital pathology is no longer viewed exclusively as a research platform or a prestige project.
This does not mean that most pathology laboratories are fully digital. The webinar repeatedly acknowledged that they are not. But “digitized” is not a simple yes-or-no category.
A laboratory may not conduct its entire primary diagnostic workflow digitally yet still own or have access to a scanner. It may be able to scan selected oncology cases, upload individual images and run a computational test through a comparatively manual process. Laboratories performing precision-oncology testing may therefore be more ready for a digital companion diagnostic than broad statistics about total laboratory digitization imply.
This distinction between full workflow digitization and test-specific digital access is important for pharmaceutical companies. A drug developer does not necessarily need every pathology case in the country to be digitized. It needs the patients being considered for its therapy to have dependable access to the required test.
Referral networks, centralized laboratories and selective scanning could provide that access while broader decentralization develops.
Reimbursement Remains the Familiar Unsolved Problem
The reimbursement discussion was one of the clearest examples of an old issue that remains old.
Laboratories can obtain CPT Category III codes associated with digitization, but coding is not the same as coverage or payment. A code may identify a service without generating incremental revenue.
In the absence of payment, laboratories must justify digitization through operational savings: remote case distribution, reduced slide handling, improved turnaround time, workforce flexibility, fewer shipping costs or greater pathologist throughput.
The panel cited health-economic work suggesting that laboratory digitization can pay for itself in approximately three to four years. But even an attractive institutional return does not necessarily solve the purchasing problem.
The department that pays for scanners may not be the department that captures the savings. Pathology, information technology, hospital administration and clinical service lines may have separate budgets and priorities. A laboratory director may want the technology but still be unable to assemble the funding, IT support and internal approvals needed to implement it.
This is a classic healthcare “wrong pocket” problem: the organization may benefit overall, but the costs and benefits fall into different pockets.
What Probably Will Not Be Reimbursed
Brown offered a relatively blunt prediction: tools that merely help pathologists work faster or improve quality are unlikely to receive separate reimbursement.
Those products may be valuable, but their business case will remain cost reduction or capacity expansion. A tool that pre-populates a synoptic report, triages cases or accelerates slide review may allow a laboratory to process 10%, 20% or even 30% more cases with the same number of pathologists. That can be economically meaningful without a new payment code.
This distinction clarifies an issue that is sometimes blurred in digital-pathology discussions:
Digitization and workflow AI primarily create operational value for the laboratory.
Predictive, prognostic and companion-diagnostic AI may create separately reimbursable medical value.
The second category has a more plausible route to payment because it can affect drug selection, diagnosis, outcomes or other measures tied directly to medical necessity.
That division is not entirely new, but the panel treated it less as a theoretical framework than as an emerging market segmentation.
The Potential Trigger: An AI-Enabled Companion Diagnostic
The most consequential “what’s new” theme was the growing expectation that computational pathology will become part of companion-diagnostic development.
Today, many pathology AI products help a pathologist perform an existing task: count cells, quantify staining, identify suspicious regions or standardize interpretation. The next generation may produce information that a pathologist cannot reliably derive by visual inspection alone.
Examples include:
identifying very low levels of biomarker expression;
characterizing spatial relationships among tumor and immune cells;
predicting response to a drug or drug class;
measuring complex tumor-microenvironment features;
combining H&E morphology with immunohistochemistry, genomics or clinical data;
and deriving prognostic or predictive signatures directly from routine tissue images.
This changes the identity of the customer and the nature of the value proposition.
A workflow tool is sold primarily to a pathology laboratory. A companion diagnostic involves the pharmaceutical manufacturer, oncologist, laboratory, regulator, payer and patient. Once a test determines whether a patient receives a high-value therapy, digitization is no longer an optional laboratory modernization project. It becomes part of the treatment pathway.
Several panelists predicted that the first significant AI-based companion diagnostic could open the market much as earlier molecular companion diagnostics accelerated adoption of genomic testing.
The reasoning is straightforward:
the drug creates demand for the test;
oncologists begin asking laboratories for the test;
pharmaceutical companies help support test availability;
laboratories acquire or access the required scanning capability;
regulators gain experience evaluating the technology;
and payers face a clearer medical-necessity argument.
The panel’s enthusiasm should be tempered. No single companion diagnostic automatically digitizes the pathology system. Pharma companies will hesitate to make a digital test mandatory if a large percentage of eligible patients cannot access it. As Brown noted, a pharmaceutical company will not want to “haircut” its drug market by 30%, 40% or 70% because the accompanying test cannot be distributed.
But this is precisely why centralized testing models may matter initially. A limited number of commercial or reference laboratories could provide access before every local pathology practice has a full digital workflow.
The likely sequence is not universal digitization followed by computational companion diagnostics. It may be the reverse: compelling companion-diagnostic use cases create islands of required digitization, which gradually connect into a broader infrastructure.
Easter compared this process to a ratchet. Once laboratories have invested in digital systems for indispensable use cases, additional digital applications become easier to adopt, and the organization is unlikely to return to purely glass-slide workflows.
Foundation Models: Important, but Not Products by Themselves
Foundation models were another genuinely current element of the discussion.
Digital-pathology algorithms were traditionally developed for relatively narrow tasks using labeled image sets: detect a tumor, score a stain or classify a specimen. Foundation models are trained on much larger and more diverse collections of images, potentially supplemented with genomic, transcriptomic, clinical and outcome data.
They provide a general representational base that can then be adapted to particular applications.
The panel was notably disciplined about what foundation models can and cannot do. A foundation model is not, by itself, a regulatory-grade clinical product. It must be fine-tuned, validated and incorporated into a defined use case.
PathAI now views its foundation model as the starting point for essentially every new product, Brown said. The company is also rebuilding older products using the foundation model as a base. Reported advantages include better robustness across staining variation, scanner types and institutions, along with improved generalization and a more scalable development process.
That may be the practical near-term significance of foundation models. They do not eliminate product development or clinical validation. They may make it faster and more successful.
The most ambitious promise is that a routine H&E slide could yield multiple clinically valuable outputs. A single image might help predict biomarker status, prognosis, drug response or features of the tumor microenvironment.
But the industry should avoid implying that the model produces every answer directly “out of the box.” The foundation model supplies the underlying architecture; individual clinical products still require task-specific development and evidence.
In other words, the foundation model is infrastructure, not the finished diagnostic.
Regulation Is Moving—but Slowly and Narrowly
Regulatory policy has advanced, although the panel’s experience suggests that the advances remain incremental.
PathAI’s AISight image-management system received a predetermined change control plan, or PCCP, as part of its FDA clearance. A PCCP allows a manufacturer to define certain anticipated product modifications and the procedures used to implement and validate them without submitting an entirely new application for every change.
In principle, this is particularly valuable for software, which evolves more frequently than conventional laboratory equipment.
In practice, Brown said pursuing the PCCP may have added at least six months to the 510(k) review. The resulting authorization allowed expansion to certain new scanner systems, but not necessarily to every seemingly modest hardware change. Adding monitors associated with already cleared systems could be allowed, while entirely new monitors remained outside the plan.
This is useful progress, but not regulatory liberation.
The broader issue is that FDA still evaluates the full “pixel pathway”: specimen preparation, staining, scanning, image handling, software analysis and reporting. A PCCP covering one component does not convert the product into free-standing software that can automatically operate with any scanner, monitor or laboratory process.
Preanalytic variability remains especially important. Tissue fixation, sectioning, staining and slide preparation can affect the image before an algorithm ever sees it. The more quantitative and subtle the algorithmic output, the more important those sources of variation may become.
The panel also noted that U.S. and European requirements are not yet sufficiently aligned. Developers may have to produce different studies or validations for different markets. Some non-U.S. systems may permit greater scanner and image-management interoperability, but local evidence and multicenter studies remain important.
Thus, regulatory change is real, but it has not yet matched the modularity that software developers would prefer.
More FDA Clearances—But Starting From a Small Base
The Digital Pathology Association’s Michael Rivers emphasized another numerical sign of momentum: the number of digital-pathology clearances has increased substantially over the past 18 to 24 months.
Philips received the first FDA clearance for a digital pathology system for primary diagnosis in 2017. Rivers estimated that the number of clearances in the most recent two-year period more than doubled the number achieved during the first six years after that milestone.
That sounds dramatic, but he added an important caution: the needle may still be near the beginning of the gauge.
This is a useful way to interpret much of the webinar. Growth rates can appear impressive because the baseline was low. More clearances, more acquisitions and more laboratory installations all represent progress, but none alone establishes that digital pathology has become the standard of care.
What is different is the density of the activity. Regulatory submissions, platform investments, pharma partnerships, acquisitions and foundation-model development are occurring at the same time.
Standardization Is Needed, but Local Evaluation Will Remain
Pathologists face a crowded algorithm market. Multiple vendors may offer products for HER2 scoring, PD-L1 scoring or similar diagnostic tasks, each reporting different performance metrics and making differently framed claims.
The panel supported greater standardization in performance reporting, potentially through target product profiles or more consistent validation frameworks.
But standardized specifications will not eliminate local evaluation.
Laboratories want to test products on their own slides, scanners, stains and patient populations. Such studies can resemble a local analytical or clinical validation. They are not conceptually mysterious, but they consume pathologist time—and evaluating ten algorithms rather than two can be burdensome.
Moreover, algorithm performance is only one part of the purchasing decision.
A slightly more accurate algorithm may lose to a competitor if it is difficult to integrate into the laboratory’s existing image-management system. Workflow design, user experience, scanner compatibility, reporting integration and technical support may outweigh small differences in headline accuracy.
This reinforces the panel’s broader ecosystem argument: the winning product may not be the algorithm with the strongest isolated performance metric. It may be the product that performs adequately while fitting most naturally into the laboratory’s end-to-end workflow.
Digital Pathology May Become Part of a Larger Multimodal System
The panel did not expect pathology AI to remain an isolated data source.
The longer-term direction is toward multimodal signatures combining morphology with genomic, transcriptomic, clinical and outcome information. Pathology may provide the starting point because nearly every solid-tumor patient already generates tissue and an H&E slide.
From that foundation, developers may add:
immunohistochemical markers;
DNA sequencing;
RNA expression;
spatial relationships;
electronic health-record variables;
prior treatment;
and longitudinal outcomes.
This has implications for companion diagnostics. Instead of a single-marker test dividing patients into positive and negative categories, future systems may identify smaller “swim lanes” of patients defined by combinations of tumor biology, morphology and molecular features.
That approach fits the evolution of oncology toward increasingly targeted drugs and combination regimens. It also raises the evidentiary bar. The more complex the signature, the more difficult it may be to explain, validate, regulate and integrate into clinical decision-making.
The panel’s answer was essentially that clinical actionability must remain the organizing principle. More information is not automatically better. The output must help determine what should happen to the patient.
A Less Discussed Frontier: Automation of the Entire Pathology Laboratory
The webinar ended with a development that was somewhat different from the usual algorithm discussion: automation of the physical anatomic-pathology laboratory.
Clinical laboratories are often highly automated. Tubes move through analyzers and tracks with limited manual handling. Anatomic pathology, by comparison, remains labor intensive. Tissue is grossed, embedded, cut, stained, transported, sorted and matched through a series of physical steps.
Rivers and Brown suggested that digitization could become the operating system around which a more automated anatomic-pathology laboratory is constructed.
The next wave may combine:
digital workflow management;
robotics;
automated tissue handling;
improved preanalytic standardization;
precision tissue extraction for sequencing;
algorithmic quality control;
and computational interpretation.
This is potentially significant because advanced AI is being attached to what Rivers called a “messy input.” A highly precise algorithm cannot fully compensate for inconsistent fixation, staining, tissue selection or slide preparation.
The future value of digital pathology may therefore extend beyond replacing a microscope with a monitor. It may enable a redesign of the laboratory itself.
So, What Is Actually New?
The panel was right that many of the issues are recurring. Reimbursement is still inadequate. Interoperability is still difficult. Most laboratories are not fully digital. Regulation still binds software to particular scanners, monitors and workflows. Laboratory capital budgeting remains fragmented.
Those are not breakthroughs.
The genuinely newer developments are more specific:
1. Consolidation has become strategic rather than speculative.
Major diagnostics and data companies are purchasing computational-pathology platforms, suggesting that the technology is becoming part of broader commercial infrastructure rather than remaining a collection of stand-alone venture-backed tools.
2. Adoption is moving into financially disciplined laboratory segments.
Reference laboratories, community systems and independent laboratories are beginning to invest, not merely academic pioneers.
3. The market is separating operational AI from medical-value AI.
Workflow tools will likely live or die on efficiency and capacity. Predictive, prognostic and companion-diagnostic tools have a more plausible route to separate reimbursement.
4. AI-enabled companion diagnostics are moving closer to commercial reality.
The panel did not present them as a distant research concept. They were described as a near-term mechanism that could connect pharma demand, oncology practice, reimbursement and laboratory digitization.
5. Foundation models are changing how products are built.
Their immediate impact is not autonomous diagnosis. It is improved robustness, generalizability and development efficiency across multiple downstream products.
6. Limited digital access may be enough to launch a precision-oncology test.
The industry may not need to wait for every laboratory to achieve comprehensive digital primary diagnosis. Central laboratories, referral systems and selective scanning can provide an intermediate distribution model.
7. The conversation is expanding from image analysis to laboratory redesign.
Robotics, preanalytics and end-to-end laboratory automation may become as consequential as the diagnostic algorithms themselves.
The Crossroads May Be a Sequence, Not a Single Moment
The webinar’s title implies a choice between paths. The panel generally rejected that framing. Digital pathology is moving forward, they said; the uncertainty is the speed and commercial structure of that movement.
A better metaphor may be a series of locks opening in sequence.
Digitization permits computational tools. Computational tools create useful workflow applications. Novel predictive tools generate medical value. Companion diagnostics bring pharmaceutical and oncologist demand. That demand strengthens the case for reimbursement and infrastructure. Broader infrastructure makes additional tools easier to deploy.
No single event completes the transition.
The most plausible catalyst is a computational pathology test that physicians believe they need, that pharmaceutical companies need distributed, that regulators have approved and that payers recognize as medically necessary. Such a product would not merely save the pathology department time. It would affect which patient receives which treatment.
Digital pathology has long had attractive technology. What it has lacked is an indispensable use case with enough clinical and economic force to pull the rest of the system along.
The panel’s underlying message was that the industry may now be approaching that use case.
The old problems have not disappeared. But for the first time, consolidation, foundation-model technology, regulatory experience, pharma interest and test-specific digital access may be aligning around a common commercial pathway. That is a more concrete claim than simply announcing another inflection point.
Whether the field has truly passed the crossroads will not be determined by the next acquisition or another impressive algorithm paper. It will be determined when a computational pathology result routinely changes treatment—and laboratories, physicians and payers have built a dependable system to deliver it.
The article is based on the webinar transcript, including the panel’s opening discussion of consolidation, predictive H&E tools, accelerated digitization and the shift from “if” digital pathology will happen to “how” it will happen.
