The Digital PATH Project: Insights from Dr. Oscar Puig

Home » The Digital PATH Project: Insights from Dr. Oscar Puig


Friends of Cancer Research kickoff meeting
Friends of Cancer Research Public Meeting: The Future of Diagnostic Tests - New Data & Modern Policy (photo courtesy of Friends of Cancer Research)
Oscar Puig, Ph.D

“Digital pathology is coming to the clinic, and it’s not an ‘if’, it’s a ‘when’” says Dr. Oscar Puig, VP of Translational Medicine & Diagnostics at Nucleai.

Nucleai, which develops AI-powered digital algorithms, is part of the Digital and Computational Pathology Tool Harmonization Project, or Digital PATH Project. Spearheaded by Friends of Cancer Research (Friends), this project brings together a consortium of academic centers, hospitals, pharmaceutical companies, algorithm developers, and advocates to address the critical challenge of variability in biomarker measurements across digital pathology platforms.

We caught up with Dr. Puig to learn about the goals, challenges, and future implications of the Digital PATH Project, and to understand how Nucleai is contributing to the endeavor. Dr. Puig’s comments have been edited for brevity and clarity.

Q:  What prompted the development of the Digital PATH Project, and what are the details of the challenge it aims to address?

The first stage of the Digital PATH Project is standardization. Validated, AI-powered algorithms, capable of clinically predictive evaluation of protein expression in images of disease tissue, do not exist. As a consequence, development teams are taking a variety of approaches toward creating digital pathology image-analysis tools. Unfortunately, if different approaches lead to different results, clinically relevant conclusions will vary. Therefore, the Digital PATH Project will first determine the variability between HER2 immunohistochemistry (IHC) biomarker measurements from different digital pathology algorithms, using the same set of breast cancer sample images.

“Different pathologists may report different scores for the same slides. Digital pathology offers the potential to analyze an image with greater accuracy and less variability.”

HER2 IHC is commonly used by pathologists to measure HER2 protein expression in tissue samples, and they know it well. An experienced pathologist can assign a score of 0, 1+, 2+, or 3+ relative HER2 expression to an IHC image in 10 seconds, hundreds of times per day. However, different pathologists may report different scores for the same slides. This is problematic because these scores are used to determine treatment paradigms.  Patients with no or low HER2 expression will be treated differently than those with high HER2 expression. Digital pathology offers the potential to quantify HER2 in an IHC image with greater accuracy, less variability, and no fatigue. Thus, after assessing variability among available algorithms and pathologist review, the Digital PATH Project will explore sources of variability and inform future approaches towards validating these type of AI-powered algorithms .

Q: Are there any objectives beyond standardization?

When pathologists score HER2 expression as positive, and as a result patients are treated with an anti-HER2 antibody, some respond to treatment and some don’t.  And patient response is not always determined by relative HER2 score. It is expected that a patient with a low (1+) HER2 score would not respond as well as a person with a higher HER2 score (2+, 3+), but that is not always the case. There is a population of patients with scores of 1+ HER2 in which the anti-HER2 antibody is very effective. Also against expectation, there is a population with scores of 3+ HER2 in which treatment with an anti-HER2 antibody is not effective.

HER2 expression scoring in breast cancer (image source)

“The next stage of this project is to identify a better way to score HER2 expression that reflects patient heterogeneity in terms of clinical response.”

Therefore, findings from this project can help identify a method of scoring HER2 expression that reflects patient heterogeneity in terms of clinical response to the anti-HER2 antibody. In this way, we hope to use AI-powered digital algorithm analysis of pathology images to identify patients most likely to respond to this therapy.

There is a great deal of information in an IHC pathology image, in addition to the level of target protein expression. There is also tumor structure, morphology, and heterogeneity. In general, this additional information is not being captured by current HER2 expression scoring, but it may be related to patient response. It is possible that developers can build a HER2 scoring algorithm that includes information beyond simply IHC staining intensity. This algorithm may better predict who will respond to anti-HER2 antibody and who will not. If so, findings from the Digital PATH Project can help inform approaches to optimize algorithms.

Q: How is Nucleai contributing to this project? 

As algorithm developers, we will provide our AI-powered digital pathology capabilities to the project. Our model will score the validation IHC images, as will the models of other algorithm developers, and together we will compare the results. We will collaborate with all members of the Digital PATH Project consortium to understand various approaches for scoring HER2 expression and factors influencing variability. It’s a collaborative effort in which we will put science first.

“Nucleai is at the forefront of developing AI-driven algorithms that can score protein expression, offering a unique approach to advancing patient care.”

The Digital PATH Project will be of value to all consortium participants. It will help physicians optimize breast cancer patient treatment paradigms. It may help pharmaceutical companies improve the efficiency of clinical trials of novel breast cancer treatments, bringing new medicines to patients faster. In addition, algorithm developers will have the opportunity to assess their models on real world data, and focus on developing robust and impactful diagnostic tools.

Nucleai is at the forefront of developing AI-driven algorithms that can score protein expression in pathology images. There are very few algorithm developers like us, so it is of value to all that we bring our skills to this consortium. And we will learn from this collaboration how to best score pathology images, what to look at, and what features to derive.

This is a new field. No one knows the correct approach. Coming at the problem from different angles and with different ways of looking at the data, and analyzing the results with agreed upon standards, is good for everybody. We will all learn together.

Q: Tell us about Friends of Cancer Research and its role in driving collaborative initiatives like the Digital PATH Project.

Friends is a nonprofit organization dedicated to expediting policy reforms, fostering groundbreaking scientific research, and facilitating the swift and safe delivery of new therapies to patients. Friend’s role as an impartial facilitator is crucial, providing a framework for meaningful cooperation among scientists, pharmaceutical entities, and policymakers. This unbiased leadership fosters trust and guides these diverse stakeholders toward collaborative success. Friends has a proven record leading collaborative projects of this kind. In a similar standardization project (ctMoniTR) they helped establish the necessary evidence to support use of circulating tumor DNA as an early endpoint for decision-making during cancer treatment.

Hillary Andrews, PhD
Director, Regulatory and Research Partnerships at Friends of Cancer Research

Nucleai recently participated in a face-to-face meeting in Washington DC to launch the Digital PATH Project.  Algorithm developers, pharmaceutical companies, and institutions providing the data sat around one table, putting aside business interests to focus on gathering scientific knowledge. Even though the concept is clear to all, each developer’s algorithm has different inputs and operates differently. We used this kick-off meeting to establish common goals and practices before the analysis begins. Our near-term objective is to evaluate initial data in June 2024.

“The number one impact is for the patients–getting the right drug to the right patient. But we also want to be more efficient in developing those drugs.”

Q: How do you envision the future applications of this project?

Drug development is time consuming and very expensive. One clear impact this project will have is to optimize the way digital spatial biomarkers are used in drug development. This optimization will improve the efficiency of pharmaceutical drug development and decrease the time required to get the right drug to the right patient.

Book A Demo

Add Your Heading Text Here