How is spatial biology used clinically?

Spatial biology reaches clinical use primarily by running on the same infrastructure as digital pathology: whole slide imaging systems, digital slide management platforms, and increasingly shared image formats. As that shared infrastructure matures, including FDA-cleared scanning systems and emerging multimodal computational models, spatial findings can move from research instrumentation toward tools a pathologist can actually use in routine practice.

How does spatial biology relate to computational pathology?

Computational pathology applies artificial intelligence and image analysis to digitized pathology images, and it is increasingly the layer where spatial biology data and routine tissue imaging converge. A 2024 foundation model called mSTAR became the first to integrate pathology slides, pathology reports, and gene expression data directly within a single model, demonstrating this convergence concretely rather than as a future goal.

Translating spatial biology into the clinic

Q: What is digital pathology?

The practice of digitizing pathology slides into whole slide images for viewing, analysis and sharing on computer systems rather than exclusively under a microscope. As of November 2024, six whole slide imaging systems had received FDA clearance for primary diagnostic use, with cleared systems achieving 92% to 96% diagnostic accuracy compared with manual microscopy in recent studies.

Highlights in this article

The research-to-clinic gap
Convergence with digital pathology
Computational pathology and AI
Clinical workflow integration
Barriers to translation

Bringing digital pathology and spatial biology together onto shared infrastructure is arguably the single development that will determine how quickly spatial insights actually reach a treating physician, rather than remaining confined to research publications. Every prior article in this cluster has addressed a piece of the pipeline, target discovery, the tumor microenvironment, biomarkers. This hub addresses the layer beneath all of them: the clinical and computational infrastructure a spatial finding ultimately has to run on to matter for an individual patient.


Key takeaways Only three AI/ML software tools for digital pathology had received FDA clearance as of a 2024 National Cancer Institute workshop, a gap the workshop attributes to limited validation datasets rather than regulatory unwillingness. Digital pathology is moving toward DICOM-based interoperability, explicitly modeled on radiology’s own successful standardization, which is the specific infrastructure convergence spatial biology stands to benefit from. A 2024 foundation model called mSTAR became the first to directly integrate pathology slides, pathology reports, and gene expression data within one computational model, demonstrating the convergence this hub describes concretely. As of November 2024, six whole slide imaging systems had FDA clearance for primary diagnostic use, with cleared systems achieving 92% to 96% diagnostic accuracy against manual microscopy. New reimbursement codes published in 2023 and 2024 show digital pathology’s financial infrastructure is being actively built, not merely discussed as a future need.

Key takeaways

Only three AI/ML software tools for digital pathology had received FDA clearance as of a 2024 National Cancer Institute workshop, a gap the workshop attributes to limited validation datasets rather than regulatory unwillingness.
Digital pathology is moving toward DICOM-based interoperability, explicitly modeled on radiology’s own successful standardization, which is the specific infrastructure convergence spatial biology stands to benefit from.
A 2024 foundation model called mSTAR became the first to directly integrate pathology slides, pathology reports, and gene expression data within one computational model, demonstrating the convergence this hub describes concretely.
As of November 2024, six whole slide imaging systems had FDA clearance for primary diagnostic use, with cleared systems achieving 92% to 96% diagnostic accuracy against manual microscopy.
New reimbursement codes published in 2023 and 2024 show digital pathology’s financial infrastructure is being actively built, not merely discussed as a future need.

The research-to-clinic gap

A finding published in a spatial biology paper and a diagnostic test a pathologist can order for an individual patient are separated by considerably more than regulatory paperwork. They are typically built on entirely different technical infrastructure: a research spatial platform often runs on specialized instrumentation and bespoke analysis software, while a hospital pathology department runs on whole slide scanners, image management systems, and reporting workflows built around entirely different assumptions and, historically, different image formats.

That infrastructure gap, not merely the regulatory validation gap covered elsewhere in this cluster, is a major reason spatial findings take years to reach clinical use even after the underlying biology is well established. A 2024 National Cancer Institute workshop on digital pathology imaging and artificial intelligence states this directly: despite digital pathology’s rapid advancement, regulatory adoption has lagged, and as of the workshop, only three AI and machine learning software tools for digital pathology had received FDA clearance. The report attributes this specifically to a validation dataset gap rather than an absence of viable regulatory pathways, which is a meaningfully different, more solvable problem than regulatory reluctance would be.

Convergence with digital pathology

The specific form this convergence is taking is standardization on shared image formats and infrastructure, rather than spatial biology building its own separate clinical ecosystem from scratch.

The same National Cancer Institute workshop report points to digital pathology’s move toward DICOM-based interoperability as the model to follow, explicitly citing radiology as the precedent: DICOM adoption in radiology enabled true enterprise-level interoperability across institutions and vendors. That same standardization path for pathology images is what would let a spatial biology finding, once validated, move across institutions and analysis platforms as readily as a radiology image already does, rather than remaining locked to whatever proprietary format a specific research instrument produced it in.

The underlying imaging infrastructure this convergence depends on is no longer experimental. As of November 2024, six whole slide imaging systems from multiple manufacturers had received FDA 510(k) clearance for primary diagnostic use, and recent clinical performance studies of these cleared systems found digital reads achieving 92% to 96% diagnostic accuracy compared with manual microscopy. Six independently cleared systems achieving that level of accuracy is evidence that digitized, computationally accessible tissue imaging is now an established clinical practice, precisely the shared surface a spatial biology application would need to run on rather than requiring its own separate clinical hardware footprint.

Computational pathology and AI

The clearest concrete demonstration that this convergence is already underway, not merely a plausible future direction, comes from a multimodal foundation model published in 2024.


One model, three modalities, at once A model called mSTAR, released in 2024 by researchers at the Hong Kong University of Science and Technology, is described as the first pathology foundation model to integrate three modalities directly within a single system: pathology slides, pathology reports, and gene expression data. The model was trained on more than 26,000 slide-level multimodal pairs drawn from more than 10,000 patients across 32 cancer types, and has been applied to tasks including metastasis identification and cancer subtyping. That combination, routine digital pathology images and gene expression data trained together within one model, is precisely the convergence this hub’s central argument describes, made concrete rather than asserted. A model of this kind does not treat spatial-adjacent molecular data as a separate system requiring its own clinical pipeline; it treats a tissue slide’s visual appearance and its underlying molecular profile as two views of the same object, which is the computational expression of the infrastructure convergence described above.

One model, three modalities, at once

A model called mSTAR, released in 2024 by researchers at the Hong Kong University of Science and Technology, is described as the first pathology foundation model to integrate three modalities directly within a single system: pathology slides, pathology reports, and gene expression data. The model was trained on more than 26,000 slide-level multimodal pairs drawn from more than 10,000 patients across 32 cancer types, and has been applied to tasks including metastasis identification and cancer subtyping.

That combination, routine digital pathology images and gene expression data trained together within one model, is precisely the convergence this hub’s central argument describes, made concrete rather than asserted. A model of this kind does not treat spatial-adjacent molecular data as a separate system requiring its own clinical pipeline; it treats a tissue slide’s visual appearance and its underlying molecular profile as two views of the same object, which is the computational expression of the infrastructure convergence described above.

Clinical workflow integration

Infrastructure and models are necessary but not sufficient; a spatial biology application also has to fit into how a pathology department is actually paid to operate, and reimbursement infrastructure is a concrete, current example of that fit being built in real time rather than left unresolved.

The American Medical Association, working with the College of American Pathologists, has published new Category III Current Procedural Terminology codes specifically for digital pathology: codes 0751T through 0763T, published in 2023 for surgical pathology services, and codes 0827T through 0856T, published in 2024, covering cytopathology, frozen section consultation, and expert consultation services. These codes exist specifically to enable remote examination of digitized slides by pathologists and the use of AI algorithms on those images. A reimbursement code is not a scientific validation, but its absence would make routine clinical use of any spatial or digital pathology workflow effectively impossible regardless of the underlying science, so its active development is a genuine, checkable sign of clinical workflow integration progressing rather than stalling.

Barriers to translation

Beyond reimbursement, a specific, checkable validation bar exists for adopting digital pathology as a clinical workflow at all, a prerequisite layer beneath any specific spatial application built on top of it.

The College of American Pathologists recommends a validation set of at least 60 cases, with concordance between whole slide image reads and glass slide reads exceeding 95%, before implementing a digital pathology workflow. That is a different, earlier validation question than the assay-specific analytical and clinical validation covered for spatial companion diagnostics in Spatial Biomarkers and Companion Diagnostics: The Next Frontier: it concerns whether a department can trust digitized slide review at all, a foundation a spatial workflow depends on rather than something a spatial assay’s own validation program can substitute for.

Three barriers recur across the evidence in this hub, worth planning around explicitly.

Regulatory clearance remains scarce relative to research output. Only three cleared AI/ML digital pathology tools, against a considerably larger body of published research, is a gap a translational program should expect to help close rather than assume someone else will.
Format and infrastructure standardization is still in progress, not finished. DICOM-based interoperability for pathology is a stated direction, following radiology’s precedent, but is not yet the settled, universal standard radiology itself achieved.
Reimbursement codes are new and still being adopted at the payer level. A published CPT code is necessary but not sufficient for routine reimbursement; payer adoption of a newly published code takes additional time this hub’s barriers should account for.

Each of the topics introduced here, the specific technical convergence between spatial biology and digital or computational pathology, deep learning applied directly to spatial tissue data, spatial biology’s use within clinical trials specifically, automating spatial workflows for pharma-scale throughput, and the regulatory path for spatial assays as diagnostics, is developed in full in its own dedicated guide within this section: Spatial Biology and Digital / Computational Pathology, AI in Spatial Pathology: Deep Learning on Tissue Data, Spatial Biology in Clinical Trials, The Path to Automating Spatial Biology for Pharma Throughput, and Regulatory Path for Spatial Assays as Diagnostics.

Broader questions of how a regulated laboratory manages compliance across digitized and computational workflows are addressed by our colleagues at Lab Manager in CLIA Compliance for Pre-Analytic, Analytic, and Post-Analytic Testing Phases. For where clinical translation sits within the full spatial biology pipeline, see Spatial Biology in Drug Discovery: From Target Discovery to Translational Medicine.

This article was produced under Drug Discovery News’s AI editorial policies.