Computational Pathology in the Era of Emerging Foundation and Agentic AI -- International Expert Perspectives on Clinical Integration and Translational Readiness

Qian Da; Yijiang Chen; Min Ju; Zheyi Ji; Albert Zhou; Wenwen Wang; Matthew A Abikenari; Philip Chikontwe

cs.CEMar 6, 2026

Computational Pathology in the Era of Emerging Foundation and Agentic AI -- International Expert Perspectives on Clinical Integration and Translational Readiness

Authors: Qian Da, Yijiang Chen, Min Ju, Zheyi Ji, Albert Zhou, Wenwen Wang, Matthew A Abikenari, Philip Chikontwe, +20 more

Organizations: Department of Pathology, Ruijin Hospital, Shanghai Jiao Tong University School of Medicine, Shanghai, China · Department of Radiation Oncology, Stanford University School of Medicine, Stanford, CA, USA · Department of Basic Education, Qingdao City University, Qingdao, Shandong, China · Department of Human Genetics, the University of Chicago, Chicago, IL, USA · Department of Computer Science, University of Warwick, Coventry UK · Department of Electrical and Computer Engineering, Carnegie Mellon University, Pittsburgh, Pennsylvania, USA · Department of Neurosurgery, Stanford University School of Medicine, Stanford, CA, USA · Department of Biomedical Informatics, Harvard Medical School, Boston, MA, USA · Department of Biomedical Data Science, Stanford University, Stanford, CA, USA · Else Kroener Fresenius Center for Digital Health, Faculty of Medicine and University Hospital, TUD Dresden University of Technology, Dresden, Germany · Department of Pathology, China-Japan Friendship Hospital, Beijing, China · Institute for Interdisciplinary Information Sciences, Tsinghua University, Beijing, China · Department of Pathology and Laboratory Medicine, Emory University School of Medicine, Atlanta, GA, USA · Wallace H. Coulter Department of Biomedical Engineering, Georgia Institute of Technology and Emory University, Atlanta, GA, USA · Ant Group · College of Biomedical Engineering, Sichuan University, Sichuan, China. · Department of Pathology, Brigham and Women’s Hospital, Boston, MA, USA. · Harvard Data Science Initiative, Harvard University, Cambridge, MA, USA. · Department of Computer Science and Engineering, The University of Texas at Arlington, Arlington, TX, USA · Department of Pathology, Division of AI and Computational Pathology, Duke University, Durham, NC, USA · Department of Medicine, Division of Nephrology, Duke University, Durham, NC, USA · Computational Pathology Group, Oncode Institute, Department of Pathology, Radboudumc, Nijmegen 6525 GA, Netherlands · Shanghai Digital Medicine Innovation Center, Ruijin Hospital Affiliated to Shanghai Jiaotong University School of Medicine, Shanghai, China. · Department of Pediatrics, the University of Chicago, Chicago, IL, USA · Comprehensive Cancer Center, The University of Chicago Medicine, Chicago, IL, USA

Abstract

Recent breakthroughs in artificial intelligence through foundation models and agents have accelerated the evolution of computational pathology. Demonstrated performance gains reported across academia in benchmarking datasets in predictive tasks such as diagnosis, prognosis, and treatment response have ignited substantial enthusiasm for clinical application. Despite this development momentum, real world adoption has lagged, as implementation faces economic, technical, and administrative challenges. Beyond existing discussions of technical architectures and comparative performance, this review considers how these emerging AI systems can be responsibly integrated into medical practice by connecting deployable clinical relevance with downstream analytical capabilities and their technical maturity, operational readiness, and economic and regulatory context. Drawing on perspectives from an international group, we provide a practical assessment of current capabilities and barriers to adoption in patient care settings.

Figures & tables

Figure 1: The evolution of multimodal pathology foundation models from passive diagnostic aids to autonomous clinical orchestrators. This framework represents a paradigm shift in computational pathology [ 26 ] , where the integration of disparate data streams [ 150 , 197 ] enables models to transcend traditional morphological assessment and ground abstract visual features in molecular truth [ 56 , 75 ] . By leveraging self-supervised alignment and structural modeling [ 29 , 34 ] , these systems bridge the gap between gigapixel-scale imaging and actionable prognostic insights, uncovering sub-visual determinants of survival and immune evasion that often escape manual evaluation [ 201 ] . The hierarchical application tiers reflect an increasing complexity of clinical utility, moving from routine phenotypic characterization toward "virtual assays" [ 21 , 185 ] that map the spatial and molecular landscape of the tumor microenvironment without additional tissue consumption. Ultimately, the transition toward agentic reasoning, facilitated by a "Supervisor-Explorer" architecture, transforms the AI from a black-box classifier into a transparent diagnostic partner that mimics expert cognitive workflows, providing an interpretable reasoning chain for complex clinical decision-making and personalized therapeutic planning [ 64 , 70 ] .

Dataset	Organs / Cancer types	Modality	Annotation type	Dataset size
Multi-organ consortia / multi-modal repositories
HTAN	Breast, Lung, Colorectal, and more (21 total)	Multiplex tissue images, Clinical data, Genomics	Various (subtype, diagnosis)	2,188 cases; 9,143 biospecimens
PLCO	Prostate, Lung, Colorectal, Ovarian	Histopathology slides, Chest X-ray, Biospecimens	Subtype, clinical data	13,165 pathology images; X-rays available
CPTAC-TCIA	Breast, Lung, Colon, and more	Histopathology slides, Radiology (CT/MR/PET)	Various (subtype, stage)	305,898 samples
TCGA	33 cancer types	Histopathology WSI, Genomics, Clinical	Various (subtype, stage)	33,179 samples
CPIA	48 organs, 100+ diseases	Histopathology images (WSI & ROI)	Subtype; multi-scale labels	21,427,877 images

Table 1: Commonly used datasets in computational pathology, grouped by primary organ/scope. Modality is specified as Whole Slide Images (WSI), microscopy images (field-of-view), or patches/ROIs to distinguish between acquisition scales. Multi-modal datasets include additional data types such as radiology, genomics, or clinical records.

Figure 2: Translational Opportunities and Barriers for Clinical Implementation of Pathology FMs . We summarize three fundamental gaps that must be bridged to translate basic pathology AI models into real-world clinical application. (1) Economic and infrastructure reality , including indirect deployment costs, large-scale data storage, long-term maintenance of imaging infrastructure, and unresolved questions regarding reimbursement and sustainability. (2) Technical gap and data challenges , arising from hardware heterogeneity, scanner- and site-specific variability, data quality and bias risks, and the intrinsic biological complexity that limits robustness and generalizability. (3) Safety, liability, and human factors , encompassing risks of generative hallucinations, cognitive automation bias, clinician deskilling, and the need for evolving accountability, regulatory, and legal frameworks for clinical adoption.

Deployment Dimension	Specific Risk	Technical Root Cause	Observable Symptom	Detection Strategy	Mitigation Strategy
Scanner variability	Feature distribution drift	Different optical pipelines	Performance drop across sites	Embedding distribution shift tests	Multi-scanner pretraining
Stain variability	Color space instability	Lab staining heterogeneity	False morphologic clustering	Stain-normalization stress tests	Stain-robust pretraining
Tissue prep artifacts	Non-biologic signal capture	Fold, chatter, necrosis	False high-confidence outputs	Artifact challenge sets	Artifact-aware augmentation
Label noise	Weak supervision contamination	Slide labels $\neq$ patch truth	Patch prediction inconsistency	Patch-slide disagreement metrics	Multi-instance training
Class imbalance	Embedding collapse	Rare class underrepresentation	Minority misclassification	Per-class calibration curves	Oversampling + metric learning
Population bias	Morphologic phenotype skew	Training cohort homogeneity	Site-specific overfit	Cross-ethnicity validation	Multi-population cohorts

Table 2: Deployment Risk and Validation Framework for FMs in Computational Pathology. This framework categorizes key barriers across deployment dimensions, identifying specific risks, technical root causes, observable symptoms, and corresponding detection and mitigation strategies.

Figure 3: Major Challenges and Translational Opportunities for Clinical Implementation of Pathology FMs We outline Six key directions for advancing the clinical translation of pathology AI FMs . (1) Bridge the gap between technical capabilities and clinical priorities to address subtle, context-dependent diagnostic needs. (2) Improve robustness to pre-analytic variability and inter-institutional heterogeneity to enhance generalizability. (3) Harness generative models for applications like virtual staining and automated reporting, while managing risks related to safety, misinformation, and accountability. (4) Develop multimodal systems that integrate histology with transcriptomics, imaging, and clinical metadata for deeper, personalized disease profiling. (5) Build scalable, cost-efficient infrastructure to support model updates and high-resolution image processing. (6) Navigate evolving ethical, legal, and regulatory frameworks to ensure diagnostic reliability and mitigate the risks of error propagation.

Figure S1 : Timeline of digital pathology milestones illustrating key developments from traditional microscopic analysis to pathology foundation models. The 1960’s marked the beginning of computerized microscopy image analysis for disease research through automated examinination of cellular structures. The 1980s saw the introduction of whole slide imaging (WSI) [ 57 ] scanners, enabling the conversion of physical slides into digital images and facilitating data storage and analysis. In the 2000s, commercial WSI scanners became widely available, with AI algorithms beginning to automate pathology tasks through digital image analysis. By the 2020s, AI-based medical devices leveraging CNNs [ 177 ] were developed, enabling diverse data type interpretation and multi-task support, further advancing the automation of disease diagnosis and clinical decision-making. The emergence of pathology foundation models represents the current frontier, moving the field toward comprehensive automation and intelligent tissue analysis.

Model	Released	Dataset Scale	Magnification	R&D Source	Architecture	Training	Hardware	Modality	Slide Enco
RetCCL	2021.Dec	32K WSIs, 15M Patches	5×,20×	TCGA,PAIP	ResNet50	MoCov2	32 V100 32G GPUs	Single Modality	no
CTransPath	2022.May	32.2k WSIs, 15.6M Tiles	20×, 40×	TCGA,PAIP	Swin Transformer	MoCoV3	48 V100 GPUs	Single Modality	no
HIPT	2022.Jun	10K WSIs, 104M Patches	20×	TCGA	ViT-S/16	DINO	–	Single Modality	no
REMEDIS	2022.Jul	29K WSIs, 50M Patches	20×	TCGA	ResNet-152(2×)	SimCLR	–	Single Modality	no
LunitDINO	2023.Apr	36K WSIs, 32M Patches	20×, 40×	TCGA, TULIP	ViT-S/(8,16)	DINO	64 V100 GPUs	Single Modality	no
Phikon	2023.Jul	6K WSIs, 43M Patches	20×	TCGA	ViT-B/16	iBOT	64 V100 GPUs	Single Modality	no

Table 3: Pathology foundation models with their key characteristics and specifications. ViT stands for Vision Transformer, with model sizes including ViT-Base (ViT-B), ViT-Large (ViT-L), ViT-Huge (ViT-H), and ViT-Giant (ViT-G), which refer to increasing parameter counts and model complexity. The table summarizes publication dates, dataset scales, magnification levels, research sources, model architectures, training strategies, and computational requirements. These models leverage advanced machine learning techniques, including self-supervised methods like DINOv2 and iBOT, as well as multimodal approaches integrating both image and text data. Dataset scales range from thousands to millions of images, with magnifications typically between 10× and 40×. Training utilizes large-scale hardware configurations including A100 and V100 GPUs to support the computational demands of these architectures.

Explore similar work

Jun 12, 2026

Democratizing and accelerating AI-driven pathology research through agentic intelligence

Computational pathology has advanced rapidly with the emergence of foundation models, yet widespread adoption remains limited by substantial technical complexity and programming requirements. Here we present PathLab, an autonomous agentic framework that translates natural-language research objectives into executable and validated computational pathology workflows through the structured composition of domain-specific skills and tools. By organizing workflow generation around reusable methodological modules, including data preprocessing, model development, evaluation and interpretation, PathLab enables studies to be specified at the level of scientific intent rather than implementation details. We evaluated PathLab across 12 public datasets spanning four representative task families: region-of-interest classification, whole-slide image classification, segmentation and survival prediction. Across all task categories, PathLab achieved non-inferior performance relative to expert implementations, while consistently enforcing semantic validity of user prompts and proactively rejecting incompatible workflow specifications prior to execution. In controlled user studies, PathLab substantially reduced the time required to generate executable analytical pipelines and enabled domain experts without programming experience to independently design, execute and evaluate computational pathology studies. Together, these results establish PathLab as a reliable interface between biomedical intent and computational execution, enabling computational pathology studies to be designed at the level of scientific questions rather than programming expertise. By lowering technical barriers to advanced AI methodologies, PathLab provides a foundation for the broader democratization of computational pathology.

Jiabo Ma, Cheng Jin, Yihui Wang +19

Department of Computer Science and Engineering, Hong Kong University of Science and Technology, Hong Kong SAR, China · Department of Pathology, Nanfang Hospital, School of Basic Medical Sciences, Southern Medical University, Guangzhou, China · Guangdong Province Key Laboratory of Molecular Tumor Pathology, Guangzhou, China +9

Computational Pathology Agentic Workflows Scientific Workflow Automation Autonomous Research Agents

Ask PDF

Aug 28, 2026

Explainable Artificial Intelligence (XAI) in Computational Pathology: Definitions, Taxonomy, and Recommendations

Computational pathology (CompPath) is transforming medicine by leveraging artificial intelligence (AI) algorithms to support diagnosis, prognosis, and treatment prediction from gigapixel whole-slide images. Clinical adoption is progressing, but is constrained by concerns about safety, accountability, and regulatory oversight in high-stakes clinical environments. Explainable AI (XAI) systems hold promise for building trust and enabling verification, yet the literature remains fragmented due to inconsistent terminology, overlapping methodological families, ad hoc validation, and current reviews. This review aims to formalize XAI methods in CompPath through the: i) introduction of a pathology-centric vocabulary comprising seven core terms; ii) development of a taxonomy across methodological families and three orthogonal axes (stage, type, scope); and iii) establishment of a task-driven framework that maps five clinical questions to recommended methods, method evaluation, and deployment context. Five key gaps between current XAI capabilities and clinical deployment are identified, and actionable steps are proposed to advance XAI for CompPath.

Shubham Innani, Suhang You, Adam Shephard +13

Division of Computational Pathology, Department of Pathology and Laboratory Medicine, Indiana University School of Medicine, Indianapolis, IN, USA · Indiana University Melvin and Bren Simon Comprehensive Cancer Center, Indianapolis, IN, USA · Tissue Image Analytics Centre, Department of Computer Science, University of Warwick, Coventry, UK +11

Explainable Medical Image Analysis Computational Pathology Explainability Evaluation

Ask PDF

Feb 25, 2026

Enabling clinical use of foundation models for computational pathology

Foundation models for computational pathology are expected to facilitate the development of high-performing, generalisable deep learning systems. However, in addition to biologically relevant features, current foundation models also capture pre-analytic and scanner-specific variation that bias the predictions made by downstream task-specific models trained on these features. Here we show that introducing novel robustness losses during downstream model training reduces sensitivity to technical variability. A purpose-designed comprehensive experimentation setup with 27,042 whole-slide images from 6,155 patients is used to train thousands of models from the features of eight well-known foundation models for computational pathology. In addition to a substantial improvement in robustness, our approach improves classification accuracy by focusing on biologically relevant features. It mitigates robustness limitations of foundation models for computational pathology without retraining the foundation models themselves, enabling development of models that are more suitable in real-world clinical use.

Audun L Henriksen, Ole-Johan Skrede, Lisa van der Schee +31

Institute for Cancer Genetics and Informatics, Oslo University Hospital, Oslo, Norway · Department of Pathology, University Medical Center Utrecht, Utrecht, The Netherlands · Department of Oncology, University of Oxford, Oxford, UK +20

Computational Pathology Pathology Foundation Models Domain Generalization Histopathology Image Classification Robust Loss Functions

Ask PDF