Dr. Catherine Guettier is an Emeritus Professor of Pathology at the University of Paris Saclay, where she was the Head of Pathology teaching for over 15 years. Dr. Guettier also led pathology departments at several Assistance Publique-Hôpitaux de Paris (APHP) hospitals in France, including Paul-Brousse, Antoine-Béclère, and Bicêtre. Between 2010 and 2012, she was the President of the French Society of Pathology. Currently, Dr. Guettier is a digital pathology consultant for Agence Régionale de Santé (ARS) in Ile-de-France.
Recently, Dr. Guettier visited Aiforia’s office in Helsinki and presented her perspectives on AI adoption in digital pathology within French hospitals. In this guest blog, she shares those reflections.
I’ve played a leading role in adopting and implementing digital pathology across hospitals in France, starting with the pilot use of digital slides for academic teaching in pathology in 2007 and, later, the first clinical use of these slides for intraoperative frozen sections (telepathology) in 2013. In 2018, our department at Bicêtre Hospital was the first in France to implement digital pathology for routine diagnosis.
Most notably, we were among the early adopters of AI in digital pathology, implementing Aiforia’s clinical prostate AI model to screen prostate biopsies first at Bicêtre and then at Saint-Louis Hospital—thanks to financial support from the Health Regional Agency Ile-de-France.
We chose Aiforia’s AI algorithm because of its performance observed while testing on our whole slide images (WSI) and because of the way the AI results are presented. This is a significant step towards accelerating the adoption of artificial intelligence in digital pathology, as it will speed up repetitive tasks in pathology workflows while improving the accuracy and consistency of image analyses.
We have found that pathologists will determine whether to adopt an AI algorithm into their workflows based on its medical and economic benefits. When choosing between different algorithms with the same or similar function, pathologists should evaluate factors such as:
When pathologists integrate AI into their diagnostic pathology routines, how can they maximize its potential?
1. Define a policy for the routine use of AI in digital pathology
Pathologists can start by evaluating whether they plan to use AI extensively or only to diagnose a few relevant indications. Likewise, it helps to understand whether these algorithms will be used in real-time or as a tool for second opinions. Interestingly, pathologists are currently uncomfortable with excluding slides analyzed as negative for cancer by AI algorithms from the pathologist’s analysis because of medical responsibility concerns.
2. Understand that AI adoption is a gradual process
AI adoption in digital pathology is still in its exploratory phase. As of 2023, fewer than 10 AI-enabled medical devices in pathology were approved by the Food and Drug Administration (FDA), unlike almost 400 in radiology, possibly due to factors like lengthy regulatory approvals and data privacy concerns.
Similarly, before 2024, the transition from conventional to digital pathology, which is the basis for AI implementation, was slow. For example, only about 20% of pathology structures in France are fully digitized, likely due to the hefty upfront costs of deployment and subsequent maintenance. A similar trend is reflected in other developed countries.
Therefore, AI is still new to many pathologists, who will gradually adapt to and trust it. Human control is easy for biomarker quantification and assistance with screening and grading, but for predictive algorithms, AI's transparency and explainability become even more important.
Today, image management systems are evolving from basic viewer interfaces to integrative platforms that offer more efficient diagnostic workflows for pathologists. This is essential because pathologists can choose the AI tools that best fit their customized workflows. For instance, Aiforia’s clinical solutions offer intelligent visualization and automated reporting, making them great candidates for pathologist workflows.
Seamless integration of AI into the laboratory infrastructure remains critical for large-scale AI adoption. From an implementation standpoint, pathologists can now choose between diagnostics-only oriented structures or structures fit for both diagnostics and research use, or for purely research-oriented work, allowing increased flexibility in workflow optimization.
Unlike AI algorithms running on centralized servers, cloud-based AI algorithms are generally easier to deploy and maintain and have lower maintenance costs. As digital pathology workflows continue to evolve with the adoption of AI, there’s much to anticipate. Faster and improved AI-driven reporting and workflow automation will enable pathologists to complete their reporting tasks more efficiently.
Looking forward, pathology AI platforms will need to be affordable, practical, interoperable, explainable, generalizable, and refundable. Today, these platforms support disease detection and feature quantification, but they will progress toward predicting outcomes and suggesting treatments. AI algorithms for future digital pathology applications will support multimodal data integration and clinical decision-making.
Read more about how Aiforia collaborates with APHP in a French article, or watch the video below: