AI Applications in Pathology

Research from Harvard & MIT used AI to unlock molecular insights in cancer pathology. Foundation models are revolutionizing computational pathology. But, most struggle to analyze entire whole-slide images (WSIs) and incorporate molecular data. 𝗧𝗛𝗥𝗘𝗔𝗗𝗦 𝗶𝗻𝘁𝗿𝗼𝗱𝘂𝗰𝗲𝘀 𝗮 𝗺𝘂𝗹𝘁𝗶𝗺𝗼𝗱𝗮𝗹 𝗳𝗼𝘂𝗻𝗱𝗮𝘁𝗶𝗼𝗻 𝗺𝗼𝗱𝗲𝗹 𝘁𝗵𝗮𝘁 𝗹𝗲𝗮𝗿𝗻𝘀 𝗳𝗿𝗼𝗺 𝗯𝗼𝘁𝗵 𝗵𝗶𝘀𝘁𝗼𝗽𝗮𝘁𝗵𝗼𝗹𝗼𝗴𝘆 𝘀𝗹𝗶𝗱𝗲𝘀 𝗮𝗻𝗱 𝗺𝗼𝗹𝗲𝗰𝘂𝗹𝗮𝗿 𝗽𝗿𝗼𝗳𝗶𝗹𝗲𝘀. • 𝗣𝗿𝗲𝘁𝗿𝗮𝗶𝗻𝗲𝗱 𝗼𝗻 𝟰𝟳,𝟭𝟳𝟭 𝗛&𝗘-𝘀𝘁𝗮𝗶𝗻𝗲𝗱 𝗪𝗦𝗜𝘀 𝘄𝗶𝘁𝗵 𝗴𝗲𝗻𝗼𝗺𝗶𝗰 𝗮𝗻𝗱 𝘁𝗿𝗮𝗻𝘀𝗰𝗿𝗶𝗽𝘁𝗼𝗺𝗶𝗰 𝗽𝗿𝗼𝗳𝗶𝗹𝗲𝘀, the largest dataset of its kind. • Enabled state-of-the-art survival prediction, identifying high-risk patients with up to 8.9% higher accuracy than previous models. • 𝗘𝘅𝗰𝗲𝗹𝗹𝗲𝗱 𝗶𝗻 𝗹𝗼𝘄-𝗱𝗮𝘁𝗮 𝘀𝗰𝗲𝗻𝗮𝗿𝗶𝗼𝘀, achieving near-clinical accuracy with just 4 training samples per class. • Introduced “molecular prompting”, allowing AI to classify cancer types and mutations without task-specific training. I like that the architecture of THREADS is notably modular. It begins with an ROI encoder based on CONCHV1.5 (a ViT-L model fine-tuned with vision–language data) to extract patch features. The patch features are then aggregated into a slide-level embedding via an attention-based multiple instance learning (ABMIL) slide encoder. In parallel, distinct encoders for transcriptomic data (a modified scGPT) and genomic data (a multi-layer perceptron) create molecular embeddings. This design not only enables integration of heterogeneous data types but also achieves remarkable parameter efficiency. For instance, THREADS is reported to be 4× smaller than PRISM and 7.5× smaller than GIGAPATH, yet outperforms them on 54 oncology tasks. Here's the awesome work: https://lnkd.in/g5y5HFuV Congrats to Faisal Mahmood, Anurag Vaidya, Andrew Zhang, Guillaume Jaume, and co! I post my takes on the latest developments in health AI – 𝗰𝗼𝗻𝗻𝗲𝗰𝘁 𝘄𝗶𝘁𝗵 𝗺𝗲 𝘁𝗼 𝘀𝘁𝗮𝘆 𝘂𝗽𝗱𝗮𝘁𝗲𝗱! Also, check out my health AI blog here: https://lnkd.in/g3nrQFxW

Processing whole slide images typically requires analyzing 18,000+ tiles and hours of computation. But what if AI could work like a pathologist? The computational bottleneck: Current AI approaches face fundamental inefficiency. Whole slide images are massive gigapixel files divided into thousands of tiles for analysis. Most systems process every tile regardless of diagnostic relevance, averaging 18,000 tiles per slide. This brute-force approach demands enormous resources and creates clinical adoption barriers. Experienced pathologists don't examine every millimeter uniformly. They strategically focus on diagnostically informative regions while quickly scanning normal tissue or artifacts. Peter Neidlinger et al. developed EAGLE (Efficient Approach for Guided Local Examination), mimicking this selective strategy. The system combines two foundation models: CHIEF for identifying regions meriting detailed analysis, and Virchow2 for extracting features from selected areas. Key metrics: - Speed: Processed slides in 2.27 seconds, reducing computation time by 99% - Accuracy: Outperformed state-of-the-art models across 31 tasks spanning four cancer types - Interpretability: Allows pathologists to validate which tiles informed decisions The authors note that "careful tile selection, slide-level encoding, and optimal magnification are pivotal for high accuracy, and combining a lightweight tile encoder for global scanning with a stronger encoder on selected regions confers marked advantage." Practical implications: This efficiency addresses multiple adoption barriers. Reduced computational requirements eliminate dependence on high-performance infrastructure, democratizing access for smaller institutions. The speed enables real-time workflows integrating into existing diagnostic routines rather than separate batch processing. Most importantly, the selective approach provides interpretability - pathologists can examine specific tissue regions influencing AI analysis, supporting validation and trust-building. Broader context: EAGLE represents a shift from computational brute force toward intelligent efficiency in medical AI. Rather than scaling hardware requirements, it scales down computational demands while improving performance. This illustrates how understanding domain expertise can inform more effective AI architectures than purely data-driven approaches. How might similar efficiency-focused approaches change AI implementation in your field? paper: https://lnkd.in/eR_Hj7ip code: https://lnkd.in/eX8wEfy6 #DigitalPathology #MedicalAI #ComputationalPathology #MachineLearning #ClinicalAI #FoundationModels

Digital pathology AI models show promise for improving lung cancer diagnosis, but robust external validation remains a critical bottleneck for clinical adoption. Methods: Researchers conducted a systematic scoping review of external validation studies for AI pathology models in lung cancer diagnosis from 2010-2024. They searched medical and engineering databases, identifying 22 studies that met inclusion criteria and assessed methodological quality using a modified QUADAS tool. Results: Key findings revealed significant validation gaps: - Only ~10% of lung cancer AI pathology papers included external validation 16 models performed subtyping tasks (LUAD vs LUSC) with AUC values ranging from 0.746-0.999 - 86% of studies showed high risk of bias in participant selection/study design -Most studies used small, retrospective datasets from single centers -Only 18% reported clinically meaningful metrics like sensitivity/specificity Conclusions: While AI models demonstrate strong performance on subtyping tasks, methodological concerns limit real-world applicability. The review highlights critical needs for: - Larger, multi-center prospective studies - Better demographic diversity in validation datasets - Standardized reporting of clinically relevant metrics - Moving beyond retrospective case-control designs - For AI pathology to achieve clinical impact, the field must prioritize rigorous external validation that reflects real-world clinical conditions and diverse patient populations. Paper and research by Soumya Arun, @Mariia Grosheva, Mark Kosenko, @Jan Lukas Robertus, Oleg Blyuss,  Judith Offman and larger team. See the comments for link to the full paper:

⚡️🔬📣 Here are our two latest preprints on how AI for Pathology can advance pre-clinical drug safety and toxicity assessment. Work led by our superstar postdoc Guillaume Jaume, Deep Learning-based Modeling for Preclinical Drug Safety Assessment 📄 Preprint: https://lnkd.in/dDsQrkfJ 🔍 Demo: https://lnkd.in/dPbKC2Xb 🌟 Insights: We trained a Vision Transformer model (TRACE) on H&E-stained whole-slide images from 150+ preclinical toxicity studies. We showed that TRACE can assist and augment pathological assessment for lesion detection, quantification and automatic dose-response characterization. TRACE was also evaluated alongside ten expert pathologists and showed better agreement with the consensus. AI-driven Discovery of Morphomolecular Signatures in Toxicology 📄 Preprint: https://lnkd.in/d-uTy9dr 🔍 Demo: https://lnkd.in/dRB-n96v 🌟Insights: We developed GEESE, an AI model trained to predict gene expression of 1,500+ targets from histology. We showed that GEESE can reveal molecular signatures associated with distinct morphologies and toxicity mechanisms that are preserved across multiple compounds and species. Congrats to Thomas Peeters, Simone de Brot, Andrew Song and everyone involved! Stay tuned for more coming soon.

AstraZeneca and Danaher Corporation launch global partnership to accelerate AI diagnostics for precision medicine: 💊The first wave of work will focus on digital and computational pathology tools, using AI to identify patients most likely to benefit from targeted therapies 💊 Leica Biosystems, Danaher’s cancer diagnostics subsidiary, will provide core technology for the alliance, supporting the entire pathology workflow 💊 AstraZeneca says diagnostic innovation is key to unlocking the full potential of next-gen medicines like antibody-drug conjugates (ADCs) 💊 The companies aim to increase global access to precision diagnostics and improve outcomes by making tests easier to scale and deploy 💊 Danaher brings expertise in digital pathology, imaging, and diagnostics, while AstraZeneca brings expertise in companion diagnostics and targeted oncology 💊 This builds on AstraZeneca’s prior collaborations with Roche, QIAGEN and Foundation Medicine to develop FDA cleared diagnostics linked to oncology drugs #digitalhealth #pharma #ai

🚀 Multimodal radiology-pathology integration for prognosis for head and neck cancer 🧬 In a recently published study in eBioMedicine by our group (and led by Dr Bolin Song) introduces SMuRF (Swin Transformer-based Multimodal Framework)—an AI-powered model that integrates CT scans and whole slide pathology images to predict survival outcomes and tumor grade in HPV-associated oropharyngeal squamous cell carcinoma (OPSCC). 🔍 Key Findings: ✅ Achieved a concordance index of 0.81 for disease-free survival prediction ✅ Outperformed unimodal models in tumor grade classification (AUC = 0.75) ✅ Demonstrates the power of multimodal deep learning for personalized cancer treatment This research highlights the future of AI-driven oncology, enabling better risk stratification and tailored treatment strategies for head and neck cancer, but has implications for other cancer as well. Read more here: https://lnkd.in/eUpFedbK #AIinHealthcare #CancerResearch #DeepLearning #MedicalAI #Oncology

#Nature: Clinical implementation of artificial-intelligence-assisted detection of breast cancer metastases in sentinel lymph nodes: the CONFIDENT-B single-center, non-randomized clinical trial Pathologists’ assessment of sentinel lymph nodes (SNs) for breast cancer (BC) metastases is a treatment-guiding yet labor-intensive and costly task because of the performance of immunohistochemistry (IHC) in morphologically negative cases. This non-randomized, single-center clinical trial (International Standard Randomized Controlled Trial Number:14323711) assessed the efficacy of an artificial intelligence (AI)-assisted workflow for detecting BC metastases in SNs while maintaining diagnostic safety standards. From September 2022 to May 2023, 190 SN specimens were consecutively enrolled and allocated biweekly to the intervention arm (n = 100) or control arm (n = 90). In both arms, digital whole-slide images of hematoxylin–eosin sections of SN specimens were assessed by an expert pathologist, who was assisted by the ‘Metastasis Detection’ app (Visiopharm) in the intervention arm. Our primary endpoint showed a significantly reduced adjusted relative risk of IHC use (0.680, 95% confidence interval: 0.347–0.878) for AI-assisted pathologists, with subsequent cost savings of ~3,000 €. Secondary endpoints showed significant time reductions and up to 30% improved sensitivity for AI-assisted pathologists. This trial demonstrates the safety and potential for cost and time savings of AI assistance. https://lnkd.in/gK3NxxuG