AI pathology transforming the possibilities in precision medicine

Published in

ZS Associates

7 min readMar 18, 2024

By: Anindita Ghosh, Khushboo Garg, and Pavan Anné (ZS Precision Medicine Center of Excellence) with contributions from Erik Walk and Michael Rivers

The expected rise in precision medicine combined with a diminishing pathology workforce warrants an investment in AI to increase workflow effectiveness. The Association of American Medical Colleges (AAMC) recently reported an 11.3% drop in the number of active pathologists between 2010 and 2015, and 63.2% of active pathologists expected to retire in the next decade. New oncology cases are expected to rise by 70% in the next two decades and the need for precision medicine in a non-oncology setting is growing in a non-linear fashion — demonstrating the business case for digitization. Excitement and expectations regarding the potential of AI in advancing precision medicine continue to build and are fueled by heightened research focus, increasing favorable attention from government and regulatory bodies and active investments from pharmaceuticals, diagnostics and technology companies. AI-based tumor pathology research publications have grown 16-fold from just 50 in 2012 to more than 800 in 2021.

At the Precision Medicine World Conference, ZS had an opportunity to garner perspectives from two esteemed leaders in the AI and digital pathology space — Eric Walk, chief medical officer, PathAI, and Michael Rivers, vice president and lifecycle leader, digital pathology at Roche Diagnostics. The panel discussion sheds light on the following:

· Exciting advancements made in recent years.

· Harnessing the potential and promise of AI.

· Barriers impacting the growth of AI in pathology.

· Need for partnerships to realize AI’s full potential.

Exciting advancements made in the recent years

The past decade has seen a host of partnerships and an outpour of significant investments from all players in the ecosystem. In the last two years alone, we saw some major pathbreaking partnerships:

· In March 2022, Roche announced a partnership with Bristol Myers Squibb (BMS) to advance personalized medicine through digital pathology solutions. Roche digital pathology is creating an AI-based image analysis algorithm to aid pathologists and generate biomarker data from clinical trial samples.

· In April 2022, GSK announced a strategic multi-year partnership to accelerate scientific research and drug development programs in oncology and non-alcoholic steatohepatitis (NASH) by leveraging PathAI’s technologies in digital pathology.

· In June 2022, Paige, a global leader in clinical AI applications in pathology, today announced a collaboration with Janssen Research & Development, LLC to evaluate the potential of a hematoxylin and eosin (H&E)-based, AI-powered biomarker test to predict the presence of certain actionable alterations in the fibroblast growth factor receptor (FGFR) genes in patients with advanced urothelial cancer, also known as bladder cancer.

· In September 2023, Microsoft announced a collaboration with Paige to build the world’s largest image-based artificial intelligence model for identifying cancer.

The journey of digital and AI in pathology has been nothing but transformational as shown in the image below. Although we’re at an early stage with widespread clinical adoption of AI and digital, there’s ubiquitous recognition of the tremendous potential this presents.

FIGURE 1: The journey of digital and AI in pathology

Harnessing the potential and promise of AI

Early in the discussion, Eric shared his perspective on the different methods that are currently leveraged in AI to aid with pathology. He cleared up a common misconception that not all digital pathology applications are driven by generative AI. This belief seems to have gained popularity because of the traction gained by ChatGPT in recent times. He emphasized the need to become fluent in different methodologies to match applications to algorithms based on the individual strength of each algorithm. To cite a few examples:

· The convolutional neural network (CNN) algorithm is highly supervised and thus provides high explainability. While it cannot be used for discovery, it’s apt for a use case like programmed death-ligand 1 (PD-L1) quantification.

· Graph neural network captures spatial relationships and hence are apt for spatial multiplex analysis.

Mike insightfully shared that the application of AI and digital will offer two-fold progress:

· Workflow benefits: In an era of pathologist workforce shortage and burnout, AI and digital can deliver lab workflow efficiencies — both time and cost-wise — through automation and standardization while also enhancing diagnosis accuracy.

· Advance patient care by influencing drug development: Potential to bring game-changing AI into the picture where there’s investment. This includes the discovery of new biomarkers by interpreting complex patterns in clinical data.

Eric concurred with the sentiment and shared that PathAI was founded on the same principle to accelerate drug development. PathAI has been working with biopharma companies to bring this vision to reality. He also added that the industry is still in the early stages of the journey and that there’s an opportunity to utilize AI to improve accuracy, reproducibility and bring continuous scoring mechanisms into play which was nearly impossible a decade ago. Spatial AI can potentially be leveraged to assess more nuanced relationships which is not possible for a human pathologist to detect. The vision of having an AI-enabled companion diagnosis (DX) is not too far and that is the vision he’s excited about.

Pharma manufacturers are setting up AI teams in their organizations. There’s a huge opportunity for pharma to proactively embed AI in clinical trials — to assist with patient recruitment, assessment of trial endpoints and enable AI-based companion diagnostics.

Figure 2: The multi-fold promise of ‘digital and AI pathology’

Barriers impacting the growth of AI in pathology

While discussing the barriers that exist today, Mike acknowledged that only 10% to 15% of pathology labs are fully digitized:

· Labs need to stand up the infrastructure and solve for massive data storage requirements and we’re seeing the industry move in the right direction to slowly eliminate these barriers — with a large number of labs initiating the process to digitize. To push digitization and automation, there’s a need to communicate the ‘what is in it for me’ to the pathologist — such as the value of going digital.

· Challenges with reimbursement of digital pathology solutions used in labs will require attention. Eric highlighted the recent approval of digital pathology codes by the American Medical Association (AMA) to be a notable step in this direction. An increasing use of these codes by pathologists will allow payers to recognize the value of digital pathology. While experiments are continuing to use federated learning, there’s no silver bullet to get around data and partnerships will be key.

· Data sharing concerns keep growing worldwide which has the potential to impact the adoption of AI. Federated learning could be an option to solve for it and is gaining traction to harness data from disparate sources in a permissible manner, but the industry is still experimenting with the concept. There’s no immediate solution to enable data sharing and interoperability, partnerships and collaborations within the healthcare ecosystem will ultimately remain key to moving the needle in this area.

Need for partnerships to realize AI’s full potential

Last month, Roche tissue diagnostics announced a key partnership with PathAI to develop AI algorithms for their companion diagnostics business. While this marks a significant milestone in the journey of AI, we need to ideate and realize many more such partnerships to realize the full potential of precision medicine. A continued fair share of commitment from every entity in the healthcare ecosystem will be necessary in overcoming the barriers in this next decade of accelerating clinical adoption of AI and digital. Partnerships that enable providers and labs with AI-driven smarter solutions, that facilitate compliant data sharing and that empower pharma and payers to bring innovative solutions to patients at a reduced cost.