Oncologists, pathologists, researchers, and industry leaders recently gathered at ASCO 2025 in Chicago to share the latest breakthroughs in cancer research and care. Among the key themes: the accelerating impact of digital pathology and AI in advancing patient outcomes and transforming R&D strategies in precision oncology. The research presented revealed a field maturing rapidly, with AI tools demonstrating transformative clinical value and practical implementation pathways. Explore our recap of key takeaways and insights.
Pathology AI Enters a New Era of Sophistication
ASCO 2025 provided compelling evidence that pathology AI has advanced substantially even in the past year. The latest technology is making significant strides in improving diagnostic accuracy, predicting treatment response, and refining risk stratification across various cancer types. Here are some of the most noteworthy findings.
Enhanced Diagnostic Precision for Targeted Therapies: Research was presented by Proscia Ready partner, Mindpeak showcasing how digital pathology-based AI can expand access to targeted therapies by improving diagnostic precision for HER2-low and ultralow scoring.1 This study involved six global academic centers and showed that AI boosted diagnostic agreement among pathologists to 86.4% (from 73.5%) for HER2-low and 80.6% (from 65.6%) for HER2-ultralow scoring. Misclassification of HER2-null cases decreased by 65%, enabling more accurate identification of patients who could potentially benefit from targeted therapies. This study highlights how incorporating digital pathology and AI into routine diagnostics has the potential to optimize treatment selection for breast cancer patients.
Predicting Risk Beyond Traditional Biomarkers: Beyond improving the accuracy of existing methods, researchers also presented the use of novel AI models as prognostic and predictive biomarkers. In stage III colon cancer, researchers from the University Medical Center Utrecht introduced the CAPAI (Combined Analysis of Pathologists and Artificial Intelligence) biomarker.2 Post-surgery circulating tumor DNA (ctDNA) is a known prognostic marker, but false negatives remain an issue. This study showed that CAPAI, an AI-driven score using H&E slides and pathological stage data, could better stratify recurrence risk even in ctDNA-negative patients.
Key findings:
- Among ctDNA-negative patients, CAPAI high-risk individuals showed 35% three-year recurrence rates versus 9% for low/intermediate-risk patients
- Over half of patients were both ctDNA-negative and CAPAI low/intermediate-risk, identifying a very low-risk group for potential therapy de-escalation
- The tool addresses false-negative ctDNA results, helping clinicians identify patients who still require intensive monitoring
Advancing Immunotherapy Patient Selection: Stanford University researchers presented work in advanced non-small cell lung cancer (NSCLC) that demonstrates how AI spatial biomarkers can predict immune checkpoint inhibitor therapy outcomes.3 Their five-feature model analyzing interactions between tumor cells, fibroblasts, T-cells, and neutrophils achieved a hazard ratio of 5.46 for progression-free survival—significantly outperforming PD-L1 tumor proportion scoring alone (HR=1.67).
This spatial analysis capability represents a paradigm shift. Rather than relying solely on protein expression levels, AI can now quantify complex cellular interactions within the tumor microenvironment, providing more nuanced predictions about treatment response.
Multimodal AI Addresses Prostate Cancer: Researchers from the University of California, San Francisco, and Artera performed external validation of a pathology-based multimodal AI (MMAI) biomarker for predicting prostate cancer outcomes after radical prostatectomy (RP).4 Using H&E images from RP specimens alongside clinical variables like age, Gleason grade, and PSA levels, the model was tested in 640 patients with a median follow-up of 11.5 years. Adjusted for the CAPRA-S clinical risk score, the MMAI score independently predicted metastasis and bone metastasis. For patients with undetectable PSA after RP, it also predicted disease progression. Those classified as RP MMAI high-risk had a significantly higher 10-year risk of metastasis (18% vs. 3% for low-risk).
This study validates the MMAI model (originally developed in biochemical recurrence patients) in post-RP patients. It demonstrates how combining image-based AI with clinical data can improve prognostic tools, guiding personalized management strategies like adjuvant therapy decisions or follow-up intensity, often with lower cost and greater accessibility than complex molecular assays.
Pioneering the Next Wave of AI Innovation in Oncology R&D
The integration of pathology AI into the oncology R&D pipeline is pivotal for identifying new drug targets, designing more efficient trials, and ensuring that the right patients are connected with the right studies, ultimately speeding the delivery of innovative therapies to patients. Specifically, pathology AI is playing a strong role in accelerating various phases of clinical trials, including improving patient recruitment and diversity, accelerating operations, and using novel trial designs with AI-based stratification.
AI-Driven Molecular Status Prediction
Johnson & Johnson Innovative Medicine’s MIA:BLC-FGFR algorithm predicts Fibroblast Growth Factor Receptor (FGFR) alterations in non-muscle invasive bladder cancer (NMIBC) patients directly from H&E-stained slides.5 Leveraging a foundation model trained on over 58,000 WSIs, the algorithm achieved 80-86% AUC, showing strong concordance with traditional testing.
The speed and efficiency of this FGFR+ detection method directly tackles NMIBC’s testing challenge: scarce tissue samples that struggle to meet the high nucleic acid requirements of traditional approaches. The algorithm’s integration into standard digital pathology workflows means oncologists can access FGFR results by testing any digitized slide from the tumor in minutes. This low-cost, accurate method could significantly enhance testing rates, accelerate treatment decisions, and optimize clinical trial enrollment for emerging FGFR-targeted therapies.
Transforming Clinical Trial Design
Daiichi Sankyo and AstraZeneca presented new data at ASCO 2025 in what is perhaps the most closely watched development in computational pathology today The presentation featured AstraZeneca’s Quantitative Continuous Scoring (QCS) computational pathology solution applied in a retrospective analysis of a NSCLC trial (TROPION-Lung02) evaluating Dato-DXd with pembrolizumab ± chemotherapy
In this exploratory analysis, QCS-positive patients in both the dual therapy and triple therapy cohorts showed a trend toward prolonged PFS compared to QCS-negative patients. Though based on smaller subsets, this first-line result aligns with the phase 3 observation that QCS enrichment identifies patients more likely to benefit, reinforcing the clinical relevance of the QCS biomarker.
Daiicho Sankyo slide in investor presentation on ASCO 2025 highlights
These data collectively support incorporating QCS into trial design and patient selection. For instance, ongoing pivotal studies (TROPION-Lung07/08) are evaluating Dato-DXd in front-line NSCLC with stratification by this AI-derived biomarker. Clinically, the ability to pinpoint “QCS-positive” tumors could help oncologists choose therapies that yield better PFS for the right patients, while sparing likely non-responders from ineffective treatment.
This research builds on QCS success in April 2025 when the U.S. FDA granted Breakthrough Device Designation to a QCS-driven companion diagnostic – the VENTANA TROP2 RxDx assay. This notably represented the first time an AI-based computational pathology device has received Breakthrough status as a cancer companion test.
The Building Blocks Behind The Breakthroughs
Behind many of these emerging AI applications lies a fundamental shift in how these algorithms learn. Foundation models—trained on vast collections of whole slide images—are becoming the backbone of digital pathology innovation.
When researchers need to develop a new AI tool for a specific cancer type or diagnostic challenge, they don’t have to start from scratch. Instead, they can build on these pre-trained models, fine-tuning them with focused datasets to tackle particular problems.
For example, Johnson & Johnson Innovative Medicine’s MIA:BLC-FGFR algorithm inputs tiles from the WSI into a Vision Transformer (ViT) foundation model to create image-based embeddings that are then used by a classification model to obtain probability for FGFR+ (see diagram here).
This approach is democratizing AI development in pathology. Where once only teams with massive resources could create effective AI tools, now researchers with smaller datasets and modest infrastructures can develop clinically useful applications. Solutions like Concentriq Embeddings are making this possible by seamlessly integrating these foundation models directly into existing R&D workflows and data systems. As a result, we’re seeing faster development cycles, more targeted solutions, and ultimately, quicker paths from laboratory innovation to patient care.
Top 5 Takeaways from ASCO 2025 in Digital Pathology & AI
- Next-Gen Therapies Demand Advanced Diagnostics. Durable responses in trials like TROPION-Lung02 underscore the need for AI-enabled digital pathology to match patients with increasingly sophisticated treatment regimens.
- Pathology AI Expands its Clinical Reach in Oncology Precision Medicine: Pathology AI demonstrated expanding clinical utility at ASCO 2025, offering risk stratification, treatment response prediction, and prognostication in oncology. Increased external validation indicated a push for robust assessment and generalizability, essential for wider adoption.
- H&E Unleashed by AI: AI is unlocking unprecedented levels of information from standard H&E-stained slides, from inferring molecular status to quantifying complex spatial TME interactions, offering the potential to reduce reliance on more costly or time-consuming tests.
- Multimodal AI and Foundation Models Signal the Future: The integration of whole slide images with clinical and genomic data, along with the rise of powerful foundation models, heralds a new era of more robust, versatile, and deeply insightful AI in oncology.
- The Ecosystem is Maturing: Increased industry collaboration, the active involvement of major technology and pharmaceutical companies with clinical institutions, and a growing emphasis on real-world validation and interoperability indicate that the digital pathology and AI in oncology is advancing towards broader, more integrated implementation to achieve precision medicine objectives.
Conclusion: The Path Forward – Realizing the Promise of AI-Powered Precision Oncology
ASCO 2025 vividly illustrated the transformative potential of digital pathology and AI in oncology. The breadth and depth of innovation presented are not just incremental improvements; they represent a fundamental shift towards a more precise, personalized, and efficient approach to cancer care. The journey from H&E slide to actionable insight is becoming faster, deeper, and more powerful thanks to these rapidly advancing technologies.
At Proscia, we are energized by these advancements and are proud to be a leader in this space. We are committed to developing and delivering enterprise-scale digital pathology solutions, like our Concentriq platform, multimodal real-world data, and global diagnostic network, that empower pathologists and scientists to harness the full potential of AI to advance precision medicine in oncology and beyond.
References
- Haab GA, Cheng C, Quang L, et al. Evaluating accuracy and concordance of pathologists and the utility of AI assistance software for digital HER2 IHC assessment in breast cancer including HER2-ultralow scoring: An international multicenter observational study. J Clin Oncol. 2025;43(suppl 16; abstr 1078).
- Franken I, Bakker M, Elias SG, et al. Complementary value of a digital pathology biomarker to post-surgery circulating tumor DNA in risk stratification of stage III colon cancer patients receiving adjuvant chemotherapy. J Clin Oncol. 2025;43(suppl 16; abstr 3604).
- Eweje F, Li R, Li Z, et al. Digital pathology-based AI spatial biomarker to predict outcomes for immune checkpoint inhibitors in advanced non-small cell lung cancer. J Clin Oncol. 2025;43(suppl 16; abstr 8569).
- Ding CKC, Shee K, Cowan JE, et al. External validation of a pathology-based multimodal artificial intelligence biomarker for predicting prostate cancer outcomes after prostatectomy. J Clin Oncol. 2025;43(suppl 16; abstr 5106).
- Ramon AJ, Koochaki F, Parmar C, et al. Inferring FGFR status from H&E images using digital pathology to identify patients for early-stage bladder cancer targeted therapies. J Clin Oncol. 2025;43(suppl 16; abstr 4593).
