Pathology is entering a new phase of digital transformation. Traditionally, pathologists have relied on glass slides, microscopes, and years of specialized expertise to examine tissue samples. Today, high-resolution digital slides combined with artificial intelligence (AI) and computational methods are creating new possibilities for analyzing tissue at scale.

This emerging field, often referred to as AI-powered pathology, brings together digital pathology, computer vision, machine learning, deep learning, natural language processing, and increasingly multimodal foundation models.

Recent research has moved beyond simply detecting cancer cells. Scientists are investigating AI systems capable of analyzing entire whole-slide images (WSIs), retrieving similar cases, generating pathology reports, predicting molecular characteristics, supporting intraoperative decisions, and connecting tissue morphology with clinical and genomic information.

A 2026 review describes advances across whole-slide imaging, AI, multimodal data integration, digital infrastructure, interoperability, federated learning, foundation models, and clinical integration.

But what exactly is changing, and how close is AI-powered pathology to routine clinical use? For broader context on AI's role in healthcare, see The Role of Artificial Intelligence in Medical Research.

Let's explore the latest developments.

What Is AI-Powered Pathology?

AI-powered pathology refers to the use of artificial intelligence and computational techniques to analyze digitized pathology specimens and support pathology-related tasks.

The process generally begins with a tissue specimen being prepared and stained. Instead of relying exclusively on a physical glass slide, the specimen can be scanned using a whole-slide imaging system.

The resulting digital image can then be processed using computational algorithms.

Depending on the application, AI may help with tasks such as:

  • Identifying and classifying tissue structures
  • Detecting suspicious or malignant regions
  • Counting cells
  • Measuring biomarkers
  • Grading tumors
  • Segmenting tissue compartments
  • Finding similar pathology cases
  • Predicting molecular or clinical characteristics
  • Supporting pathology report generation
  • Assisting research and drug development

Importantly, AI does not eliminate the need for pathology expertise. Current research is largely focused on developing systems that can support pathologists, researchers, and clinical workflows.

Digital Pathology: The Foundation of AI-Based Analysis

AI-powered pathology depends heavily on digital pathology infrastructure.

Digital pathology involves scanning pathology slides into high-resolution digital files that can be viewed, transferred, stored, and computationally analyzed. The FDA describes digital pathology as involving the scanning, visualization, analysis, transfer, storage, and interpretation of tissue information in digital form.

Whole-slide imaging is particularly important because pathology specimens can contain enormous amounts of visual information.

A single digital slide may contain millions or billions of pixels. This creates both an opportunity and a technical challenge.

Traditional computer vision approaches often analyze smaller image regions, known as patches or regions of interest. Newer computational pathology systems increasingly attempt to understand information across the entire slide.

That shift is important because many diagnostic patterns depend on relationships between different areas of tissue rather than an isolated microscopic region. For related insights on diagnostic advances, see Pathology and Laboratory Medicine: Diagnostic Advances and Biomarker Discovery.

Medical Journal IMJ Health Call for Papers

The Rise of Pathology Foundation Models

One of the biggest developments in computational pathology is the emergence of foundation models.

Foundation models are large AI models trained on extensive datasets so that their learned representations can potentially be adapted to multiple downstream tasks.

In pathology, researchers are developing models specifically designed to understand histopathology images and, increasingly, relationships between images and text.

The potential advantage is significant. Instead of training a separate AI model from scratch for every pathology task, researchers can start with a pretrained foundation model and adapt it to a particular application.

Recent studies show how quickly this area is developing.

TITAN: Connecting Whole-Slide Images and Text

A 2025 study published in Nature Medicine introduced TITAN, a multimodal whole-slide foundation model designed to connect pathology images with textual information.

The model was pretrained using more than 335,000 whole-slide images alongside pathology reports and synthetic captions. Researchers evaluated it on tasks including rare disease retrieval, cancer prognosis, cross-modal retrieval, and pathology report generation.

This illustrates an important change in computational pathology. AI systems are increasingly being designed not just to recognize visual patterns but to connect what appears in tissue with how those findings are described clinically.

Multimodal AI in Pathology

Multimodal AI combines different types of information rather than analyzing one data source in isolation.

In pathology, this may include:

  • Whole-slide images
  • Pathology reports
  • Clinical information
  • Genomic data
  • Radiology information
  • Laboratory results
  • Molecular biomarkers

The goal is to create systems capable of understanding relationships between these different sources.

This could eventually support more comprehensive clinical research and decision-support tools.

However, multimodal pathology AI remains an active research area. The availability, quality, privacy, and standardization of clinical datasets continue to influence how effectively these models can be developed and validated.

Generative AI Meets Whole-Slide Imaging

Generative AI has also begun moving into computational pathology.

One recent example is SlideChat, a multimodal generative AI system designed to work with whole-slide pathology images rather than only small image patches.

Published in Nature Cancer in September 2026, the research describes a system trained using 274,233 multimodal instruction samples. The researchers evaluated it across multiple cohorts covering 31 cancer types and reported improvements over the evaluated baseline systems on closed-ended questions and report-generation tasks.

This research points toward a potential future in which pathologists could interact with pathology AI using natural-language questions.

For example, rather than manually searching through thousands of image regions, an AI assistant could potentially help retrieve relevant regions, summarize visual findings, or assist with report-related tasks.

Such capabilities remain subject to clinical validation, workflow testing, regulatory requirements, and careful evaluation of errors.

AI for Intraoperative Pathology

Another important research direction is the use of AI during surgery.

Intraoperative pathology can help surgeons make decisions while an operation is underway. However, frozen-section interpretation can be challenging because tissue preparation and interpretation must happen quickly.

A September 2026 Nature Medicine study introduced CRISP, a clinically oriented foundation model developed using more than 100,000 frozen sections from 10 medical centers. Researchers evaluated it on more than 15,000 intraoperative slides and subsequently assessed it in a prospective cohort involving more than 3,000 patients.

The researchers reported that the system maintained performance across multiple institutions, tumor types, and anatomical sites and investigated its role in supporting surgical decisions.

The study is notable because it moves beyond retrospective image benchmarks toward prospective clinical evaluation.

That distinction matters. An AI model performing well on a curated research dataset is not automatically equivalent to an AI system being reliable in everyday clinical practice.

Whole-Slide Image Retrieval Is Becoming More Sophisticated

Another emerging application is content-based image retrieval.

Imagine a pathologist examining an unusual tumor and wanting to find visually similar cases from a large digital pathology archive.

AI-based retrieval systems can potentially search large collections of slides and identify images with similar visual characteristics.

Recent 2026 research evaluated pathology foundation models for whole-slide image retrieval across 9,387 diagnostic slides covering 17 organs and 60 diagnoses. The researchers found substantial variation between organs and diagnoses and noted that some rare or closely related subtypes remained difficult to retrieve reliably.

This is an important finding because it demonstrates both the promise and the limitations of foundation models.

AI can help search enormous image collections, but difficult diagnostic distinctions remain challenging.

Foundation Models Are Not Automatically Reliable

The rapid growth of pathology foundation models has created an important question:

Does a larger or newer AI model necessarily perform better?

Recent benchmarking research suggests that the answer is not that simple.

A 2026 Nature Communications study benchmarked 32 foundation models across 41 pathology tasks and more than 17,500 whole-slide images. The researchers found that model performance varied according to the task and dataset, while model size and pretraining dataset scale did not consistently predict downstream performance.

This has practical implications.

A pathology AI system should not be evaluated simply by asking how large its training dataset was or how sophisticated its architecture sounds.

Instead, researchers need to examine:

  • External validation
  • Generalization across institutions
  • Performance across tissue types
  • Scanner variability
  • Rare disease performance
  • Reproducibility
  • Clinical usefulness
  • Error patterns
  • Human-AI interaction

Scanner Differences and AI Robustness

Digital pathology introduces a challenge that does not always receive enough attention: the same tissue can look different depending on how it is digitized.

Different scanners, staining protocols, image-processing pipelines, and laboratory practices can introduce variations.

A 2026 study examining foundation models for histopathology image retrieval specifically investigated scanner-related variation. The researchers found that some vision-transformer-based foundation models demonstrated strong performance and generalization across scanner conditions, while also emphasizing the importance of understanding covariate bias.

This is important for real-world deployment.

An AI model trained primarily using images from one environment may not automatically behave the same way when used with slides produced by another laboratory or scanner.

AI and Computational Pathology for Precision Medicine

One of the most exciting areas of computational pathology is precision medicine.

Pathology images contain information about:

  • Tumor architecture
  • Cell morphology
  • Nuclear characteristics
  • Immune-cell patterns
  • Tissue organization
  • Tumor microenvironment

Researchers are investigating whether AI can extract features from these images that correlate with molecular alterations, prognosis, treatment response, or other clinically relevant characteristics.

Foundation models may accelerate this research because a single learned representation can potentially be adapted to multiple downstream tasks.

However, these applications require rigorous validation before they can be relied upon for clinical decision-making. For related insights, see Single-Cell Multiomics: Breakthrough Discoveries and Latest Research in Cancer Heterogeneity.

Federated Learning and Privacy-Preserving Research

Medical AI requires large datasets, but pathology images are linked to patient information and are therefore subject to privacy and governance requirements.

One approach receiving attention is federated learning.

Instead of transferring all patient data into one central database, federated approaches can allow participating institutions to train or evaluate models while keeping sensitive data within their respective environments.

Recent reviews identify federated learning and large-scale data repositories as important developments in the evolution of digital and computational pathology.

This could become increasingly important as researchers seek larger and more diverse datasets without unnecessarily centralizing sensitive patient information.

The Role of AI in Pathology Workflow Automation

AI-powered pathology isn't limited to diagnosis.

Computational systems can potentially assist with workflow-intensive tasks such as:

  • Slide prioritization
  • Case triage
  • Image quality assessment
  • Cell counting
  • Tissue segmentation
  • Biomarker quantification
  • Image retrieval
  • Report assistance
  • Quality control
  • Research data annotation

Automation of repetitive activities could allow pathology professionals to spend more time on complex cases and clinical interpretation.

However, workflow automation needs to be designed around actual laboratory processes. An algorithm that performs well technically may provide little practical value if it creates additional review steps or integrates poorly with existing systems.

Regulatory and Validation Challenges

The development of AI-powered pathology also raises regulatory questions.

Digital pathology itself has been an area of regulatory research for years. The FDA has identified challenges involving whole-slide imaging performance, interoperability, image quality, AI reproducibility, and assessment of generalizability.

For AI pathology systems, validation needs to consider more than algorithmic accuracy.

Important questions include:

Does the model work on data from different institutions?

Does it perform consistently across scanners and staining protocols?

What happens when the model encounters an unusual case?

Can pathologists understand when the system may be uncertain?

Does AI actually improve workflow or patient care?

These questions are especially important as pathology AI moves from research environments toward clinical applications. For related insights on AI diagnostics and accountability, see The Black Box of AI Diagnostics: Liability Gaps, Legal Challenges, and Future Accountability Models.

Why Human Expertise Still Matters

AI can analyze images at extraordinary scale, but pathology is more than image classification.

Clinical interpretation can involve:

  • Patient history
  • Specimen type
  • Previous diagnoses
  • Laboratory findings
  • Imaging
  • Molecular results
  • Treatment history
  • Clinical context

A pathology AI system may identify an important visual pattern without understanding the complete clinical situation.

For this reason, human-AI collaboration is an important theme in current research.

The objective is increasingly not simply to replace human interpretation but to explore ways AI can provide useful information while allowing qualified professionals to review, contextualize, and act on the findings.

Key Challenges Facing AI-Powered Pathology

Despite rapid progress, several barriers remain.

1. Data Diversity

AI models need representative data from different populations, institutions, scanners, staining methods, and disease presentations.

2. Rare Diseases

Rare cancers and unusual pathological patterns can be difficult to learn because training datasets may contain relatively few examples.

3. Generalization

A model that performs well on one dataset may perform differently on another. Recent benchmarking studies demonstrate that model performance can shift depending on the dataset, task, and tissue type.

4. Explainability

Clinicians may need to understand why an AI system produced a particular output, especially when the result could influence a clinical decision.

5. Integration

AI needs to work with laboratory information systems, digital slide viewers, reporting platforms, and existing clinical workflows.

6. Regulation and Accountability

Clear processes are needed for validation, monitoring, updates, error management, and responsibility when AI-assisted systems are used clinically.

What Could the Future of Computational Pathology Look Like?

The future is likely to involve increasingly integrated pathology platforms rather than isolated AI algorithms.

A future pathology workflow could potentially combine:

Whole-slide imaging + foundation models + clinical data + molecular information + natural-language interfaces

Such systems could help professionals search large image archives, identify potentially important regions, quantify biomarkers, retrieve similar cases, summarize findings, and connect morphological observations with other clinical information.

The development of multimodal and whole-slide models suggests that pathology AI is moving toward systems capable of handling increasingly complex forms of information.

But technological capability should not be confused with clinical readiness.

The strongest research direction is likely to be one that combines advanced AI with rigorous validation, transparent evaluation, appropriate human oversight, and carefully designed clinical workflows.

AI-powered pathology is developing rapidly. Digital pathology has created the infrastructure for tissue to become computationally accessible, while deep learning and foundation models are providing increasingly sophisticated methods for analyzing that information.

Recent research has expanded beyond basic image classification toward whole-slide foundation models, multimodal AI, generative pathology assistants, image retrieval, intraoperative decision support, and precision-medicine applications.

At the same time, current research makes clear that significant challenges remain. Generalization, scanner variation, rare diseases, dataset diversity, clinical validation, privacy, interoperability, and regulatory requirements all influence whether an AI system can move successfully from a research paper into routine practice.

The future of pathology is therefore unlikely to be simply "AI versus pathologists." A more realistic direction is pathologists working with increasingly capable computational systems, combining human clinical expertise with the speed, scale, and pattern-recognition capabilities of AI.

As research continues, the most meaningful advances will be those that demonstrate not only impressive technical performance but also reliable, reproducible, and clinically useful outcomes. For those considering doctoral research in this area, Top 10 Pharmaceutical Research Topics for PhD offers guidance on selecting impactful research directions.

Continue Exploring Our Blog