Report from the WG Digital & Computational Pathology (DCP):

What happened in Digital and Computational Pathology at the European Congress of Pathology, Basel, 2022?


The WG Digital & Computational Pathology, formerly called the WG IT (Computational) enjoyed an active presence at the 34th European Congress of Pathology (ECP) 2022 in Basel. Whether tackling the important question of “AI: useful or useless” in different disease entities, to a live demonstration of “AI in the wild” during the hugely popular Computational One-Day Symposium, or discussing critical aspects of quality management and ethics at the AI Taskforce meeting, the presence of digital and computational pathology could certainly be felt. Importantly, abstracts for poster sessions and oral free papers were abundant, highlighting the increasing relevance and interest of these topics for pathologists across Europe and beyond. Here, we give a short report on the activities connected to digital & computational pathology at the ECP.

Keywords: Digital pathology, computational pathology, artificial intelligence

Joint Symposium Breast Pathology & IT in Pathology

The first joint symposium took place on the opening day, Sunday September 4th, 2022. The session was chaired by Simone Münst and Antonio Polonia who moderated not only excellent speakers’ presentations but also lively discussions. The talk by Balazs Acs set the stage for AI in the field of breast pathology, giving an overview of the current perception and reality of computational solutions and challenges. The open question “Is AI ready for prime time?” was addressed, highlighting challenges in the field. This was followed by a presentation by Inti Zlobec and Zsuzsanna Varga, who presented a comparison of machine learning algorithm performance, from various open-source and commercial platforms and scanners, highlighting that all tools have challenges on the same images. In a second step, one such algorithm was applied to the IBCSG 8/9 clinical trials and compared with the original manual scoring. Discrepancies were discussed. This was followed up by a talk by Francesco Ciompi who spoke about the state-of-the-art in breast cancer diagnosis using AI approaches. Several important features such as mitosis and grading were highlighted along with the challenges of generating such algorithms. Paul van Diest, a pioneer in the digital diagnostic lab transformation, showed processes, pitfalls and IT architecture, showcasing image analysis integration and possibilities with the image management systems and viewers. Another hotly discussed presentation was that of Francesco Martino, speaking about internet security in digital pathology. Legal and regulatory aspects were debated, as well as cloud-solutions for storage and digital pathology applications. This session triggered a large interest in the audience. Antonio Polonia showed and discussed the most recent studies predicting molecular biomarkers (ER, PR and HER2) in breast cancer using HE stain and deep learning approaches, highlighting the HEROHE Grand Challenge. Both benefits and difficulties of implementation of these methods in clinical practice were discussed. Finally, Vilja Pietiäinen spoke on the topic of precision medicine in solid tumors using AI and patient-derived cells, still at the research level though.

Keynote lecture

The first keynote of the ECP was held by Inti Zlobec with the title “2001: A Digital Pathology Odyssey”. Directly inspired by the Stanley Kubrick classic film “2001: A Space Odyssey”, the presentation themed around space, told the story of a pathologist, Dana, thrust into the world of digital pathology not by choice but by necessity during the Covid-19 pandemic. In this new era, she learns how to navigate new topics: from going digital in the lab and IT basics, to developing AI algorithms with different strategies for deep learning, to the challenges of interoperability and limitations from the lack of structured data but aided by the resources and current initiatives around the globe engaging in these topics. Novel technologies, including spatial transcriptomics and high-dimensional protein analysis (multiplexing) were discussed and their incredible visual relationship to stars and galaxies in space. As future perspectives, 3D imaging and multi-modal analysis of different data types were mentioned. The take home message is: we are all learning together, and we are not alone in our digital transformation. The beauty of the pathology of today is in the interdisciplinary collaborations, which will lead to innovative solutions for patients and is a critical pillar of personalized health.

Joint Symposium Digestive Diseases Pathology (GI) & IT in Pathology

The question addressed in the GI session, chaired by Xuchen Zhang and Inti Zlobec was “Artificial intelligence in digestive diseases: needed or needless?”. The first talk was by Luca di Tommaso, who put the current state-of the-art into perspective in colorectal and liver diseases. This was followed by Sara Pires Oliveira, who described her deep learning studies on colorectal polyp classification and validation, addressing the question whether or not computer-aided diagnostics are ready for clinical acceptance. Iris Nagtegaal gave an overview of the current colorectal cancer studies utilizing AI, including microsatellite instability (MSI) prediction from the H&E stains, and going beyond 2D, with a discussion on 3D imaging for tumor budding and pseudobudding classification. Going deeper into molecular classifications, Viktor Koelzer spoke of his work using the image Consensus Molecular Subtypes, intratumoral heterogeneity and prediction of prognosis based on H&E staining in colorectal cancers. Moving to the upper gastrointestinal tract, Heike Grabsch spoke on the application of deep learning based image analysis to support diagnostic decisions in oesophageal and gastric cancer. Overall there are very few studies compared to colorectal cancer, some promising results, but extensive validation in independent datasets is needed before implementation into the routine workflow might be considered. Finally, Rupert Langer turned to inflammatory bowel disease. He showcased that most image analysis work being performed relates to endoscopic images with pathology being used as a “gold standard”, rather than to histopathology images utilized for AI predictions, themselves. Taken together, AI in digestive diseases is also in its early days, but there is a lot of room to answer interesting and diagnostically relevant problems.

Joint Symposium Pathology in Favour of Developing Countries & IT in Pathology

Joining Forces! This was the theme and the spirit in which our third joint session took place.  Chaired by Giulia di Falco, Antonio Polonia and Inti Zlobec, the speakers covered a range of exciting and different topics showcasing their motivation for working in the field of AI in pathology and the impact of the digital transition in different parts of the world. The first speaker Nina Linder gave an inspiring account of how AI can be used for cytopathology diagnosis in Africa, giving details on her study for cervical cancer screening and readiness for the clinic. The next speaker was Catarina Eloy, who demonstrated her own institute’s telepathology model and benefits in remote parts of Portugal. Peter Fritz highlighted the applications of his AI models and approach to help diagnose different disease entities in a pathologist-barren area of Afghanistan. Coming from India, Rajiv Kumar Kaushal presented a highly relatable depiction of how his hospital is preparing for the digital transition, underscored by advantages and limitations. Finally, Heather Dawson shared experiences in digital pathology from her institute, and a special use case of the “patho-mobile”, namely a frozen-section-specific vehicle used to perform intraoperative diagnoses at different hospitals in Switzerland. This session truly underlined the common points in digital pathology transition across geographical borders.

Computational Pathology Symposium 2022

ECP Basel 2022 was the host for the 7th computational pathology symposium. The CP symposium aims to be a forum for discussion and exchange of knowledge in the area of computational pathology, by bringing together key speakers from both academia and industry to facilitate such interaction in the broadest possible manner. The morning session was moderated by Anna Bodén (Linköping, Sweden). We dedicated the first hour of the symposium to discussing the concept of grand challenges in computational pathology. Grand challenges have proven to be among the most powerful and effective driving forces in our field [1]. Geert Litjens (Radboudumc, Nijmegen, Netherlands) provided an overview of grand challenges, showing how these competitions have radically improved the quality of solutions in fields as diverse as self-driving cars, netflix movie recommendations, and medical imaging. The basic idea of a challenge is to set up an international competition, in which interested researchers all work on the same problem using the same data and being judged using the same criteria. As a result, the solutions that the AI researchers produce are perfectly comparable, leading to valuable insights into which approaches are most appropriate for the problem at hand. Typically, by combining a number of the top-performing AI models, even better solutions can be produced. Well-known challenges in histopathology include AMIDA, GLAS, CAMELYON, PANDA, TUPAC, and MIDOG. In the second presentation, Nasir Rajpoot (Warwick, UK) explained the setup of the CONIC22 challenge, in which two tasks in H&E tissue sections of colon are defined: 1. segmentation and classification of cell nuclei, and 2. assessment of the cellular composition of the tissue. To develop these methods, a very large set of 500,000 annotated nuclei was provided as training data. Even though the results of the challenge were very promising, and outperform previous solutions, there is still considerable room for improvement for these tasks. In the third presentation, Jeroen van der Laak (Radboudumc, Nijmegen, Netherlands) presented results from the recent TIGER challenge. In TIGER, participants were required to produce AI to segment tumor and tumor stroma in breast cancer H&E sections, detect individual lymphocytes within the tumor stroma (so-called tumor infiltrating lymphocytes, TILs) and assess the TIL density. TILs have been shown to be prognostic as well as predictive for specific subtypes of breast cancer. In this challenge, two leaderboards were used: one to assess the quality of the different AI solutions (segmentation of different tissues, detection of lymphocytes) while the second expressed the prognostic value of the submitted AI (on the basis of the c-index of a multivariable Cox regression model, predicting recurrence free survival). One special feature of TIGER is the fact that data from clinical trials could be used (which cannot be shared directly with participants) by having participants upload their containerized AI solutions, allowing the challenge organizers to perform the evaluation without having to disclose the images.

In the second session, Maria Gabrani (IBM, Zürich, Switzerland) showed how we can build AI models, without losing explainability (which is a typical complaint about deep learning algorithms). Using the motto ‘we’re giving pathologists a Ferrari, while they're used to ride a bicycle’, she expressed the need to improve our understanding of the data, using multimodal data integration. Her work applies graph neural networks (GNNs), in which relationships between tissue components are studied [2]. These relationships can be translated into concepts that pathologists can relate to (e. g., cellularity). Next, Henning Muller (HES-SO, Sierre, Switzerland) explained how, in the EU-Examode project, fully automated analysis of free-text pathology reports was developed. The analysis results in concepts, which are subsequently used as weak labels with corresponding whole-slide-images to train deep learning models. This fully automated pipeline allowed to use extreme-scale data for AI training, as it does not rely on any human involvement in the AI development [3]. Furthermore, also Henning expressed the importance of interpretability, arguing that the impact of wrong decisions in medical diagnostics can be very high. He showed how regression concept vectors can give a better understanding of what AI models 'look at' to arrive at an outcome. Henning: 'Interpretability of deep learning is key for integration of tools into clinical workflows’. Last in this session, Dimitris Metaxas (Rutgers University, New Jersey) first showed strategies to reduce the workload for pathologists to produce annotations for AI training [4]. He showed how such strategies can be used to predict gene mutations (e. g., TP53 mutations) in breast cancer on the basis of H&E images. An ensemble of different GNNs was used to study relationships between different kinds of tissue features, aiming to predict genetic mutations and survival for colorectal cancer directly from the images [5].

In the afternoon session, chaired by Shan Raza (University of Warwick, UK), top-ranked submitted abstracts were presented. Oliver Lester Saldanha presented his talk ‘Swarm learning for decentralized deep learning in gastric cancer histopathology’, which was awarded the Computational Pathology Symposium award 2022 for the highest ranking abstract. Also in this session: Niclas Blessin - BLEACH&STAIN, a novel multiplex fluorescence immunohistochemistry framework that facilitates a fast high throughput analysis of >15 biomarkers in more than 3000 human carcinomas; Julie Swillens - Pathologists’ first perspectives on barriers and facilitators of computational pathology implementation in histopathology; Maschenka Balkenhol - The prognostic value of deep learning based mitotic count for breast cancer molecular subtypes; Fabio Pagni - Pathologist validation of a machine learned biomarker for risk stratification in colon cancer; John Connelly - Predicting genetic variation from quantitative tissue phenotypes using explainable machine learning; Muhammad Aslam - Successful deployment of an AI solution for primary diagnosis of prostate biopsies in clinical practice.

In the last, interactive session, results were presented of a crowd-sourcing experiment that was performed in the weeks before the ECP, coordinated by Khrystyna Faryna and Leslie Tessier (Radboudumc, Nijmegen, Netherlands). In this experiment, pathologists were broadly invited (through LinkedIn, and direct contact) to upload digitized prostate biopsy images, which were subsequently analyzed by the top 5 AI models from the PANDA grand challenge [6]. As an introduction to this session, Geert Litjens (Radboudumc, Nijmegen, Netherlands) presented the PANDA challenge, which was aimed at developing AI solutions for Gleason grading of prostate cancer.

Next to the top 5 PANDA algorithms, PAIGE (New York, NY), a company dedicated to providing AI solutions to the Pathology field, participated in the session. Additionally, a panel of pathologists was invited to give their Gleason score for the uploaded biopsies, using remote diagnostics through a web-based interface. In the interactive session, which was moderated by Inti Zlobec (Bern University, Bern, Switzerland) and Jeroen van der Laak (Radboudumc, Nijmegen, Netherlands, Linköping University, Sweden), Rainer Grobholz (Kantonsspital Aarau, Switzerland) presented a number of cases in detail, asking the audience to provide their Gleason score for each case using their mobile phones. Scores of the audience could be compared with the panel of pathologists, as well as with the AI models. With this setup, pathologists in the audience could directly experience how their opinion compared to the scores of their colleagues (underlining the inter-pathologist variability), as well to the AI models. It also highlighted the power of AI in analyzing cases from very diverse sources.

AI in Pathology Task Force Session

The session was chaired by Jeroen van der Laak (Radboudumc, Nijmegen, Netherlands, Linköping University, Sweden) and Darren Treanor (Leeds Teaching Hospitals NHS Trust and Linköping University). The task force chair Jeroen van der Laak welcomed the audience and outlined the reason for the session to establish a task force for artificial intelligence in pathology on behalf of the ESP. The specific objectives for the session were: 1. To engage the community in the new taskforce; 2. To identify relevant topics for the taskforce to address; 3. To make a plan for next steps. A quick survey learned that the audience (consisting of 86 people from approximately 18 countries) consisted of approximately 50 % pathologists, and 30 % engineers/ scientists.

In this session, two topics which are considered of high relevance for the adoption of AI in pathology were discussed. David Brettle (University of Leeds, UK) gave an introduction on the importance of quality and quality assurance in digital pathology and AI. The presentation gave rise to a long discussion about the role of quality management. Some key points raised were:

  • That quality is important for digital pathology and AI but more importantly for the pathology process as a whole (including pre-digitisation steps and the pathologist’s opinion) – i. e., that quality in digital pathology is not in isolation of general quality management
  • Several attendees noted that digitisation enhanced the interest in quality in their lab, either in order to make scanners work, or that the digitisation drew attention to quality issues
  • Some discussion and disagreement about whether AI could or could not overcome variation in image stain or quality
  • he role of regulatory mechanisms alongside the pathology profession and the AI manufacturers in jointly ensuring good results was highlighted

Next, Rouven Porz (University Hospital Bern, Switzerland) gave a presentation on several (high level) ethical questions regarding AI in pathology. He provided very useful recommendations and highlighted the need for the experts (the ESP community) to lead the discussion and educational activities in this area. The following discussion included the following points:

  • The misconceptions about AI, for example that AI is intelligent and independent
  • The importance of understanding and addressing the issue of bias in AI datasets – with a warning from Dr Porz that completely eliminating all bias in any field is very challenging
  • The opportunity that DP/ AI present to address global health inequalities
  • The role of the human pathologist in the loop of decision making
  • The need to highlight the benefits that AI can provide
  • The importance of aligning with existing initiatives (e. g., the WHO guidance on governance of AI for healthcare)

Next, a discussion on potential topics for the task force to address resulted in this list:

  1. Quality assurance
  2. Ethics
  3. Validation and verification of digital pathology and AI
  4. Regulatory questions around DP/ AI (e. g. the upcoming changes to IVDR)
  5. Training of pathologists (both residents and fully qualified) in digital pathology and AI
  6. The role of open science in digital pathology and AI
  7. How to engage the pathology community in identifying the medical problems / grand challenges to address


ECP Basel 2022 was a fruitful experience, bringing together numerous collaborators in the field of Digital and Computational Pathology. The activities connected to the WG Digital & Computational Pathology and to the AI Taskforce will continue to increase in frequency and in importance. We hope to organize an exciting ECP Dublin in 2023.


Prof. Dr. Inti Zlobec
Institute of Tissue Medicine and Pathology (ITMP), University of Bern, Switzerland
Dr. Antonio Polonia
Institute of Pathology and Molecular Immunology, University of Porto, Portugal
Prof. Dr. Darren Treanor
Leeds Teaching Hospitals NHS Trust, UK and Linköping University, Linköping, Sweden
Prof. Dr. Jeroen van der Laak
Department of Pathology, Radboud University Medical Center, Nijmegen, The Netherlands
Center for Medical Image Science and Visualization, Linköping University, Linköping, Sweden