Advancements in Pathology Through Transformative Technologies

Today in Cell, we published new research showing how AI can help accelerate cancer discovery. With GigaTIME, we can now simulate spatial proteomics from routine pathology slides, enabling population-scale analysis of tumor microenvironments across dozens of cancer types and hundreds of subtypes.   Developed in partnership with Providence and the University of Washington, our hope is that this work helps scientists move faster from data to insight, revealing new links between genetic mutations, immune activity, and clinical outcomes, and ultimately improving health for people everywhere. https://lnkd.in/dSpPdtzz

Roche buys PathAI in big bet on AI-native diagnostics infrastructure: 🔘Roche is acquiring AI-powered digital pathology company PathAI in a deal worth up to $1.05bn, signalling continued pharma confidence that AI will become deeply embedded in diagnostics infrastructure, not just drug discovery and clinical development 🔘The deal builds on a partnership that started in 2021 and expanded into AI-enabled companion diagnostics in 2024, showing how many of these acquisitions are increasingly the result of long courtships rather than sudden AI shopping sprees 🔘PathAI focuses on digital pathology, where tissue samples are converted into high resolution digital images that AI can help analyse. The goal is not just speed and efficiency, but more precise and reproducible cancer diagnostics and treatment selection 🔘Roche appears to be making a broader strategic bet that pathology data will become a foundational layer for precision medicine, biomarker discovery, clinical trials and AI-enabled companion diagnostics, particularly in oncology 🔘One of the bigger themes here is the shift from AI as “software on top” toward AI-native workflows. Digital pathology potentially transforms pathology from a largely manual microscope-based process into a scalable data and analytics platform 🔘More broadly, pharma increasingly seems willing to buy AI capability rather than simply partner around it, with AstraZeneca and Modella AI a recent example of players trying to internalise AI, data and programmable biology capabilities rather than sit one step removed from them 💬It is another sign that future competitive advantage may not just come from owning drugs, but from owning the data, workflow and decision infrastructure around them #digitalhealth #ai #pharma

We’re entering an era where tissue biomarkers won’t be measured by eye. Even when they’re IHC-based, the next generation of assays won’t rely on pathologists estimating intensity or proportion scores like we do for HER2 or PD-L1. That’s one clear takeaway from Pathology Visions 2025 (Digital Pathology Association) AstraZeneca’s QCS system for TROP2 was a standout. The FDA now recognizes its “normalized membrane ratio” (NMR) as a novel biomarker—a computational measurement that quantifies membrane and cytoplasmic expression patterns linked to drug internalization and efficacy. It’s literally impossible to score an NMR by eye. This marks a turning point. In AstraZeneca’s model, deep-learning systems (ART and SSTC) handle the region and subcellular segmentation. The locked, reproducible NMR outputs are the biomarker. The pathologist’s role shifts to quality assurance—verifying the image quality, overlays, and workflow integrity—not subjective interpretation. This is what a “human-in-the-loop” system really means. We also saw new data from Tamara Jamaspishvili using PathAI‘s AIM-PD-L1 algorithm in NSCLC: AI-derived PD-L1 quantification (AIM-PD-L1) stratified survival better than human pathologists. In fact, AI reclassified nearly 30% of cases—and those reclassifications correlated more closely with objective response and overall survival. We’ve known that PD-L1 interpretation by pathologists is inconsistent. Now we can measure how inconsistent. More biomarkers will follow—HER2 may be next. The most powerful biomarkers ahead may not be visible at all. They’ll be defined by quantitative patterns that correlate with mechanism of action and clinical benefit, not by thresholds that a human pathologist can eyeball. The pathologist’s role isn’t disappearing—it’s evolving. From measuring by eye to governing the quality, validity, and interpretability of algorithm-driven diagnostics. As AI learns to quantify what the human eye can’t perceive, diagnostic pathology is shifting—from seeing to measuring. #DigitalPathology #Pathology #ArtificialIntelligence #PrecisionOncology #PathVisions25 #AIMedicine #Cancer #biomarkers

Computational pathology has officially moved from a "nice-to-have" to a "must-have." No longer is it an optional upgrade to the workflow; rather, it is the essential infrastructure for modern drug discovery and clinical development. As a surgical pathologist by training, I remember vividly the days of painstakingly counting positive cells, annotating IHC slides by hand and spending hours under a microscope to identify the best avenues for targeting an individual’s cancer. Today, AI and computer vision have radically transformed the approach to discover, develop and deliver biomarkers for patient selection and treatment decision making. The paradigm shift is from simply "reading" slides to "analyzing" data. Driven by state-of-the-art domain-specific foundation models, we can now understand tumor biology in ways that significantly improve treatment predictions. This is now a core component of patient selection strategies for both single therapies and combination regimens. I recently joined Rob Monroe, MD, PhD from Leica Biosystems and Karan Arora from Danaher Corporation for a discussion with The Pathologist on how we are using these digital ecosystems to break through traditional bottlenecks. To realize the impact of AI today, an ecosystem of partnerships is essential. No single stakeholder can solve this alone. We ought to collaborate and bring together domain knowledge, technical expertise and the infrastructure required to transform the patient journey. Read part one of our discussion here: https://lnkd.in/diawUFr3 #DigitalPathology #AI #PrecisionMedicine #Biomarkers #Oncology #DrugDiscovery #ClinicalDevelopment #DomainSpecificSuperIntelligence

For years, AI in pathology has primarily focused on narrow tasks such as classification, segmentation, and detection. While these approaches have shown tremendous value, they often depend heavily on large annotated datasets and predefined labels. Vision-Language Models (VLMs) introduce a fundamentally different paradigm. By combining image understanding with natural language reasoning, VLMs can interpret histopathology images in a far more contextual and human-like manner. Instead of simply identifying pixels or regions, these models can begin to describe morphological patterns, correlate findings, summarize observations, and support pathologists with explainable insights. Potential applications in Digital Pathology include: - Automated interpretation of histologic patterns - AI-assisted generation of descriptive findings - Context-aware triaging of whole slide images - Natural language querying of pathology datasets - Enhanced explainability of AI outputs - Multi-modal integration of pathology, genomics, and clinical data - Educational and training support for pathology workflows The future of computational pathology is likely not just “seeing” images, but understanding and communicating pathology in clinically meaningful language. We are entering the era of multimodal pathology AI. #DigitalPathology #ComputationalPathology #ArtificialIntelligence #VisionLanguageModels #Pathology #MachineLearning #HealthcareAI #GenerativeAI #MultimodalAI #CancerResearch #DrugDevelopment

Recent advances in machine learning inform precision medicine and translational research. We introduce a pioneering approach that integrates pathology imaging with transcriptomics and proteomics to identify predictive histology features associated with critical clinical outcomes in cancer. We utilize 2,755 H&E-stained histopathological slides from 657 patients across 6 cancer types from CPTAC. Our models effectively recapitulate distinctions readily made by human pathologists: tumor vs. normal (AUROC = 0.995) and tissue-of-origin (AUROC = 0.979). We further investigate predictive power on tasks not normally performed from H&E alone, including TP53 prediction and pathologic stage. Importantly, we describe predictive morphologies not previously utilized in a clinical setting. The incorporation of transcriptomics and proteomics identifies pathway-level signatures and cellular processes driving predictive histology features. Model generalizability and interpretability is confirmed using TCGA. We propose a classification system for these tasks, and suggest potential clinical applications for this integrated human and machine learning approach. A publicly available web-based platform implements these models. Interesting paper from the Clinical Proteomic Tumor Analysis Consortium: https://lnkd.in/eAArJwDv

For years, advancing pathology AI meant one thing: hoarding massive, proprietary datasets to train ever-larger models. But a new paper suggests the field might be hitting diminishing returns on simply adding more data. Histopathology data is highly redundant. The vast majority of tissue patches look similar, creating a long-tailed distribution where rare, diagnostically critical features get drowned out by common morphologies. Consequently, brute-force scaling is highly inefficient. Saarthak Kapse et al. released 𝙂𝙚𝙣𝘽𝙞𝙤-𝙋𝙖𝙩𝙝𝙁𝙈, a 1.1 billion parameter open-weight foundation model that challenges the "more is better" dogma. Despite using only 10% to 20% of the training data required by current leading proprietary models (like Virchow2 or UNI2), it achieves state-of-the-art results across clinical, molecular, and robustness benchmarks. Here is how they achieved unprecedented data efficiency: • 𝘼𝙪𝙩𝙤𝙢𝙖𝙩𝙚𝙙 𝘿𝙖𝙩𝙖 𝘾𝙪𝙧𝙖𝙩𝙞𝙤𝙣: Instead of using every available WSI, the team built a fully unsupervised pipeline using hierarchical clustering and stratified sampling to select tiles based on morphological diversity. This approach filters out redundant patterns, allowing the model to train on just 177k public WSIs while prioritizing high-entropy content like rare histological variants. • 𝙏𝙝𝙚 𝙅𝙀𝘿𝙄 (𝙅𝙀𝙋𝘼 + 𝘿𝙄𝙉𝙊) 𝙋𝙧𝙚𝙩𝙧𝙖𝙞𝙣𝙞𝙣𝙜 𝙍𝙚𝙘𝙞𝙥𝙚: Standard DINO pretraining captures good global morphology, but the authors went a step further by introducing a novel dual-stage strategy. After an initial DINO phase, they froze the encoder and used it as a teacher for a JEPA-based student. By tasking the student with predicting visible regions and outpainting missing ones, the model learned highly fine-grained, spatially-aware representations without relying on raw pixel reconstruction. • 𝙎𝙩𝙖𝙩𝙚-𝙤𝙛-𝙩𝙝𝙚-𝘼𝙧𝙩 𝙍𝙤𝙗𝙪𝙨𝙩𝙣𝙚𝙨𝙨: When tested on the PathoROB benchmark (which measures resilience to varying scanners and stains across multi-center datasets), 𝙂𝙚𝙣𝘽𝙞𝙤-𝙋𝙖𝙩𝙝𝙁𝙈 established a new state-of-the-art average Robustness Index of 0.888, significantly outperforming much larger, data-heavy models like Virchow2 and UNI2. 𝙏𝙝𝙚 𝙩𝙖𝙠𝙚𝙖𝙬𝙖𝙮: The future of clinical AI is not just about who has the biggest private dataset. Intelligent data curation and optimized learning objectives can match or exceed the performance of unconstrained scaling, offering a path to more accessible, transparent, and robust pathology AI. https://lnkd.in/evdMNVak --- Keeping up with the literature is increasingly a team sport. This analysis was supported by NotebookLM and grounded in my own review and experience. If you found this useful, let me know in the comments. If it missed the mark, I want that feedback too. Weekly briefings on making vision AI work in the real world → https://lnkd.in/guekaSPf

When we talk about AI in biomedicine, it is easy to focus on molecules, proteins, or genes. But one of the most vital and challenging frontiers is digital pathology. Every cancer diagnosis in the world still begins with a pathologist looking down a microscope at a stained tissue sample. Today, those slides can be digitized into whole slide images (WSIs), gigapixel files that capture every cell in astonishing detail. And that is where the challenge begins. 📍 WSIs are enormous. They are far larger than the images standard computer vision models were built for. Training AI to handle them requires tiling, stitching, and multiscale modeling just to make the data tractable. 📍 The biological complexity is staggering. A single slide may contain millions of cells of dozens of different types, some of which (like neutrophils or eosinophils) are extremely rare and understudied. 📍 Noise and variability are the norm. Different hospitals use different stains, scanners, and protocols. Robust AI has to generalize across all of it if it is to be useful in the clinic. That is why OWKIN’s release of HistoPLUS is so important. 🔬 HistoPLUS is not just another pathology model. It sets a new state of the art for nuclei detection, segmentation, and classification across 13 cell types. It achieves 5% better detection and 24% higher F1 classification than prior methods, while using 5× fewer parameters. 🌍 Most importantly, it generalizes across multiple cancer types, which is a crucial step if AI is to support pathologists in real-world hospitals rather than controlled research settings. And in the spirit of open science, Owkin has released both the weights and the implementation openly, so that researchers, clinicians, and startups can build on this work. By releasing HistoPLUS on Hugging Face, they maximize its impact, reach, and adoption, which means it can accelerate scientific achievement even further. It is a reminder that open-source AI in medicine is not just about transparency. It is about accelerating discovery, democratizing access, and ultimately improving patient care. 👏 Kudos to OWKIN (especially Lucie Gillet, Benjamin Adjadj, Pierre-Antoine Bannier, and Guillaume Horent) for pushing the field forward and for sharing tools that the entire community can build upon. 📄 Preprint: https://lnkd.in/ev-jpXj4 🤗 Hugging Face: https://lnkd.in/eReMnECX

Pathology is finally becoming a measurable and standardized data generator. In our new commentary paper, we discuss the shift from traditional morphologic interpretation toward AI-enabled integration of morphology, molecular biology, and clinical data. We also highlight an important point for industry and healthcare systems: value comes from validated tools that address real workflow needs and biomarker challenges. Next steps to be made: • scalable quantitative biomarker assessment • AI pre-screening and triage for molecular testing • digital endpoints for clinical trials • greater standardization across laboratories • faster translation of pathology data into clinical evidence The next phase of precision oncology will increasingly depend on how effectively pathology data are converted into actionable insight. Link to the full article: https://lnkd.in/dP4iPkCx #Pathology #ComputationalPathology #DigitalPathology #AI #PrecisionOncology #Biomarkers #ClinicalTrials #HealthcareInnovation Konstantinos Venetis, PhD IEO Istituto Europeo di Oncologia Università degli Studi di Milano

Microsoft Research Pathology-AI Breakthrough https://lnkd.in/gaiERFUV GigaTIME uses AI to translate routine Hematoxylin and Eosin (H&E) pathology slides directly into virtual multiplex immunofluorescence (mIF) images, effectively mapping the spatial proteomic landscape of the Tumor Immune Microenvironment (TIME). • How it Works: The framework was trained on an immense dataset of 40 million cells with paired H&E and mIF data across 21 proteins, allowing it to bridge cellular morphology and protein states. • Scale and Impact: We applied GigaTIME to a massive virtual population dataset—14,256 patients from 51 hospitals, generating 299,376 virtual mIF slides spanning 24 cancer types. • Clinical Discovery: This population-scale analysis, previously impossible due to mIF data scarcity, uncovered 1,234 statistically significant associations linking proteins, biomarkers, staging, and survival. • Future of Pathology: GigaTIME enables large-scale clinical discovery and patient stratification directly from routine, cost-effective H&E slides, revealing new spatial and combinatorial protein activation patterns. This work marks a significant step by Microsoft toward democratizing high-resolution TIME analysis and accelerating the development of precision cancer therapies. #Microsoft #AIinMedicine #Pathology #CancerResearch #SpatialProteomics #PrecisionMedicine #GigaTIME Figure Courtesy: Cell