AI solutions drive digital pathology transformation: Insights from ASCO, ECDP, RCPath conference

Sparked by recent advances of artificial intelligence applications to healthcare, we explored the challenges and opportunities of digital pathology transformation at several international conferences this month.

Pathology is facing two major challenges: a global shortage of pathologists as case load increases, and recent breakthroughs in oncology, driven by new biomarker discovery, which require pathologists to provide ever more precise and complex diagnostic analysis. Digital transformation to the pathology ecosystem heralds an opportunity to meet these challenges directly, with multiple AI solutions introduced to the market to support and streamline pathologists workflows. The curiosity and demand for digital transformation and the integration of AI tools is growing; catalyzing the innovation of solutions ranging from quantification tasks, to biomarker screening, and outcome prediction.  

Prediction: All pathology labs will be digitized by 2030³

In discussion at the European Congress of Digital Pathology (ECDP), Andrey Bychkov shared the interesting cross-continent study he conducted with his team in “Exploring the adoption of digital pathology in clinical settings”. Their findings indicate that while digital pathology is still the future for many, it is already the present for some. Costs are a major issue, but as technology advances and vendors compromise, they anticipate that equipment will become more affordable and interoperable. This decade may be a time when anatomic pathology finally embraces the digital revolution on a large scale.

Highlights from the study:

• Large laboratories are adopting digital pathology more than smaller ones
• Complete switchover to digital is still rare today
• Many initial concerns did not materialize after implementation (such as increased turnaround time, computer crashes, or the perceived superiority of the microscope).
• Most non-digital laboratories plan to go digital by 2030

‍

Will future pathologists be ready for the digital revolution?

Although the future of digital pathology is bright, teaching digital pathology skills to future pathologists is still inconsistent. The Royal College of Pathology (RCPath) symposium provided insights into the impact of the “digital revolution” on how future pathologists will be required to learn and adapt. It also highlighted the disparities between digital learning and microscopic learning within the same country (UK), but also across nations

• Pathologists are well aware of the ongoing “Digital Revolution” and they are actively working on adapting their training to stay aligned with this transformation and smoothly manage the transition from “standard pathology” to “digital pathology”.
• During the next few years, we anticipate a mismatch between pathologists trained to work in a digital environment vs. on a microscope and fully digitalized lab vs, traditional lab.
• In addition to the impact on clinical routine, pathologists are aware that they need to understand how AI works and how it is validated, to understand how the tool can best be used in their workflow to optimize their workload.
• After focusing on the digital workflow and AI tools impact, the training sessions dealt with equity, ethical and environmental issues raised by this digital revolution: bias in data (ethnicity, geographical, area of expertise), equal access to better care for minority ethnic groups, privacy/consent and data protection, data ownership, environmental impact–critical issues that Owkin is exploring in our ongoing research.

What are some of the breakthroughs being made today?

A few examples of promising research and development shared at this month’s many oncology and pathology focused conferences:

• At ASCO, Dr. Andrew J. Armstrong presented the results of an AI-derived digital pathology-based biomarker to predict the benefit of long-term ADT (Androgen Deprivation Therapy) in addition to radiotherapy in men with localized, high-risk prostate cancer. Such findings are very promising as it would radically improve clinical management of a subset of prostate cancer patients not expressing the biomarker. They could avoid side effects associated with long-term hormone therapy as they would not benefit from it, compared with biomarker-positive patients. These results await further validation before they can be transposed into clinical practice, potentially paving the way for faster administration of the right treatment to the right patient.
  

• Gain of function mutations of the FGFR3 gene have been reported to be essential driver mutations in a subset of muscle-invasive bladder urothelial cancer (MIBC). At ASCO, Owkin researchers presented affordable and fast pre-screening solutions such as artificial intelligence based image classifiers that could help to reduce turnover times of FGFR3 mutational testing and save costs arising from universal mutational testing. The study proposes an AI-based approach to identify MIBC FGFR3 mutated patients on routine histological slides. By ruling out 57% of MIBC FGFR3 wild type tumors with high sensitivity, this model is the first to reach valuable performance for clinical use. Consequently, this study is an encouraging  development of FGFR3 pre-screening tools to ease the management of patients with MIBC - facilitating access to promising immunotherapies for FGFR3-mutant metastatic MIBC patients.
  
  
• At ECDP “Introducing ChatGPT to Image Classification for Histopathology” poster explored how Large Language Models (LLM) could be used to help image classification in histopathology. ChatGPT is an AI model that uses generative pre-training transformer (GPT), and it has been optimized using supervised and reinforcement learning. Open AI offers the ability to fine-tune GPT models to perform specific tasks. The objective of this research is to use these models to classify images. Will LLM find their way into histopathological classification and even diagnosis?

Want to learn more about ChatGPT and how LLMs might be leveraged in the life science space? Read our blog, Alban fact checks ChatGPT’s views on how generative AI will impact drug discovery.

Owkin’s role in shaping the future of digital pathology

Eager to meet this transformation and ensuing complexities head on, Owkin is committed to developing robust AI diagnostics to not only support pathologists in their day to day, but also to improve patient outcomes. Working closely with our extensive academic network, our aim is to provide high-performance digital pathology solutions that make precision medicine more accessible to more patients. To discover how these solutions are integrated in clinical routine, read our recent news about MSIntuit CRC’s deployment into one of France’s largest pathology networks. Connect with us as we continue to pursue this mission.

References

1. 1. Amstrong A. AI-Derived Digital Pathology Biomarker Predicts Benefit of Long-Term ADT with Radiotherapy in Localized High-Risk Prostate Cancer. ASCO, June 2023.
2. Charlie Saillard, Pierre-Antoine Bannier, Philipp Mann, Charles Maussion, Christian Matek, Arndt Hartmann, and Markus Eckstein, AI-based identification of FGFR3 mutation status from routine histology slides of muscle-invasive bladder cancer. Journal of Clinical Oncology 2023
3. Daniel Gomes Pinto, Andrey Bychkov, Naoko Tsuyama, Junya Fukuoka, Catarina Eloy Exploring the adoption of digital pathology in clinical settings - Insights from a cross-continent study, ECDP, June 2023.