AI is changing how pathology is practiced by giving pathologists tools that improve accuracy, reduce variation, standardize results, speed up workflows, and boost performance. In general, AI in pathology is used in six key areas. Knowing how these types of algorithms work helps pathologists use AI more effectively in daily practice – leading to better care for patients.
Here, we examine the six essential pillars of AI that should be in every pathologist’s lexicon and consider some case studies of laboratory process improvements. Finally, we detail the particular promise of generative AI in pathology practice.
Detection
Detection algorithms help pathologists by automatically spotting specific features or abnormalities in pathology images. These tools reduce the chance of missing subtle signs of disease, improving overall accuracy. One strong example is the CONFIDENT-B trial, which showed how AI can successfully help detect breast cancer spread in sentinel lymph nodes. Another tool, the AIM-PD-L1 algorithm, can reliably detect metastatic cancer in lymph nodes for non-small cell lung cancer (NSCLC), helping reduce false negatives and differences between pathologists’ interpretations.
Classification
Classification algorithms help pathologists tell the difference between disease subtypes by analyzing patterns in tissue appearance and markers. These AI tools make it easier to accurately identify tumor types, which supports more personalized treatment decisions.
In our own research, we found that using whole slide imaging (WSI) and AI in soft tissue pathology can improve diagnostic accuracy and help differentiate between benign and malignant tumors. We see AI as a valuable support tool that can enhance the pathologist’s ability to make accurate diagnoses.
Quantification
Quantification algorithms help pathologists measure biomarker levels more accurately and consistently than traditional visual methods. This reliable, standardized approach is key for making informed treatment decisions in precision cancer care. AI tools, like those presented at the 2025 ASCO Annual Meeting, have shown they can improve the accuracy of identifying HER2-low and HER2-ultralow breast cancers – subtle but important subtypes that may respond to new targeted therapies.
Prognosis
Prognostic algorithms use tissue features to predict how a disease might progress, offering insights that go beyond standard staging and grading. These AI tools help doctors make better decisions about treatment and follow-up care.
For example, the ArteraAI Prostate Biomarker Test combines digital pathology with clinical data to predict how a patient will respond to therapy and what their outcome might be. It can assess how aggressive the cancer is and estimate the risk of it spreading or coming back, providing a risk score. This helps doctors tailor care to each patient, which can improve both survival and quality of life in prostate cancer.
Prediction
Predictive algorithms help determine how likely a patient is to respond to a specific treatment, allowing for more personalized care. These tools combine different types of data – like imaging, tissue features, molecular markers, and clinical information – to guide treatment choices. The ArteraAI Prostate Biomarker Test, for example, can predict whether a patient will benefit from therapies like androgen deprivation therapy (ADT). Its ability to both predict outcomes and guide treatment makes it a powerful tool for tailoring prostate cancer care.
Workflow & quality improvement
AI-driven workflow and quality improvement algorithms streamline laboratory processes, ensuring higher efficiency and diagnostic consistency. They automate routine tasks, cut down on variation, and improve overall performance and productivity – saving time and reducing costs. With the adoption of digital pathology workflows, the practice of pathology has evolved. To keep pace with the advancements, it’s essential to leverage AI tools to improve both efficiency and quality.
Seeing improvements
Through the application of these six algorithm types, AI is playing a pivotal role in enhancing quality assurance and streamlining workflows in pathology laboratories. Even gold standard methods can be improved when AI algorithms are introduced.
For instance, immunohistochemistry (IHC) staining is fundamental in pathology for visualizing specific tissue components. However, variability in staining quality can lead to diagnostic inconsistencies. A study using AI to help detect breast cancer spread in sentinel lymph nodes showed that pathologists using AI were less likely to rely on IHC, reducing its use by about 32 percent and saving around €3,000 per case. The AI-assisted approach also saved time and improved diagnostic sensitivity by up to 30 percent – highlighting AI’s safety and its potential to save both time and costs.
Improving precision pathology also depends on better standardization of tissue biomarker staining. IHC tests have had error rates up to 10 times higher than other clinical tests, increasing the risk of incorrect diagnoses. As a result, there’s growing demand for stricter regulations to ensure quality.
Accuracy is also essential for biomarker testing in precision oncology, especially when determining HER2 status in breast cancer. However, traditional manual scoring of HER2 can vary between pathologists and may lead to inconsistent results – particularly when distinguishing low and ultra-low HER2 levels. AI-based tools offer a promising solution by delivering more precise, consistent, and reproducible assessments.
In one study, an AI system matched expert pathologists with over 92 percent accuracy and helped improve agreement between pathologists – especially in challenging cases like differentiating HER2 0 from 1+.
AI has also enabled subcellular-level analysis, detecting ultra-low HER2 expression in some cases previously labeled as HER2 0. In one analysis, AI reclassified nearly a quarter of such cases, potentially expanding access to targeted treatments for more patients.
Beyond HER2, AI-based systems have improved accuracy and speed in classifying multiple breast cancer biomarkers, including estrogen and progesterone receptors. Deep learning models have shown strong performance in categorizing HER2 status from IHC-stained images, helping to streamline interpretation and reduce variability across observers.
These advances demonstrate how AI can enhance diagnostic accuracy, reduce variability, and ultimately broaden access to personalized treatments – key goals in modern breast cancer care and precision oncology.
Looking ahead
Generative AI (GenAI) is set to transform pathology by improving diagnostic accuracy, simplifying workflows, and enhancing educational tools. Its growing range of uses is reshaping traditional practices and helping raise standards in patient care.
One major application is the creation of synthetic data to train AI models. GenAI can produce diverse, representative datasets that help overcome the limitations of small or unbalanced real-world data. This not only improves model performance and reduces bias but also protects patient privacy, as the synthetic data are not tied to real individuals.
GenAI can also assist with drafting diagnostic reports. By converting spoken observations into clear, structured reports, it helps reduce manual errors and speeds up documentation. This allows pathologists to spend more time on clinical decision-making and less on administrative tasks.
Another strength of GenAI is its ability to combine different types of medical data – such as tissue images, genetic information, and clinical history – into a single analysis. This comprehensive view helps uncover patterns that may be missed when examining each data type separately, supporting more accurate diagnoses and better personalized treatment plans. Recent developments include AI tools that can interpret pathology images alongside text-based data to assist with complex cases.
As GenAI becomes more common in pathology, experts and organizations have begun establishing guidelines to ensure its ethical and effective use. These efforts focus on best practices, regulatory standards, and real-world examples, helping ensure that GenAI advances support patient-centered, responsible care.
Overall, it's clear that pathology services can benefit from AI assistance in multiple forms, from combining gold standard processes, to GenAI approaches to diagnostics. There’s always something new on the horizon, aiming to improve patient care services.
Marilyn Bui is Senior Member, Professor of Pathology, and Scientific Director of the Analytic Microscopy Core at Moffitt Cancer Center & Research Institute, in Tampa, FL USA
Ghulam Rasool is Assistant Member of Machine Learning at Moffitt Cancer Center & Research Institute, in Tampa, FL USA
Newsletters
Receive the latest pathologist news, personalities, education, and career development – weekly to your inbox.
