Eric Walk Of PathAI: How AI Is Disrupting Our Industry, and What We Can Do About It

Professional development — pathologists need to learn about AI methods and be fluent enough in the language of AI to effectively select the best solutions to drive growth and value in their specific practice setting.

Artificial Intelligence is no longer the future; it is the present. It’s reshaping landscapes, altering industries, and transforming the way we live and work. With its rapid advancement, AI is causing disruption — for better or worse — in every field imaginable. While it promises efficiency and growth, it also brings challenges and uncertainties that professionals and businesses must navigate. What can one do to pivot if AI is disrupting their industry? As part of this series, we had the pleasure of interviewing Dr. Eric Walk.

Dr. Eric Walk is Chief Medical Officer at PathAI in Boston, MA. He is head of the medical group, overseeing the pathology team at the PathAI Diagnostics laboratory in Memphis, TN. Dr. Walk has over 20 years of experience in precision medicine, oncology drug development, and companion diagnostics development. Prior to joining PathAI, he was Chief Medical and Scientific Officer at Roche Tissue Diagnostics/Ventana Medical Systems, where he led medical and scientific affairs, overseeing the development and FDA clearance/approval of over 20 510k and PMA IVD assays, companion diagnostics and digital pathology algorithms. Dr. Walk began his industry career at Novartis Oncology, where he held positions in early clinical development and translational medicine, working to implement biomarker and precision medicine strategies for early and late stage targeted oncology therapeutics. Dr. Walk is a Phi Beta Kappa graduate of Johns Hopkins University and holds a MD from the University of Virginia School of Medicine. He is board certified in Anatomic and Clinical Pathology and is a Fellow of the College of American Pathologists (CAP). He currently is a member of the CAP Digital & Computational Pathology Committee.

Thank you so much for joining us in this interview series. Before we dive into our discussion our readers would love to “get to know you” a bit better. Can you share with us the backstory about what brought you to your specific career path?

I’ve actually had a very non-traditional career as a physician. It’s been a journey I never would have predicted, but one that’s been incredibly fulfilling.

After medical school, I did my residency and fellowship training in pathology because I was interested in disease mechanisms and wanted to focus on the then nascent field of precision medicine. After practicing pathology for 2 years in the hospital environment, I transitioned into the pharmaceutical industry with Novartis in 2002 and had the privilege to work on the development of some of the first precision medicine therapies in oncology. That led me to the field of diagnostics, joining Ventana / Roche Tissue Diagnostics in 2005 and later becoming the Chief Medical Officer (CMO), leading the medical and scientific affairs team and helping to develop cutting edge tissue diagnostics, including multiple companion diagnostics in partnership with pharmaceutical companies.

In 2021, I realized the tremendous potential of digital and AI pathology to transform the field and patient care more broadly, joining PathAI as the company’s first CMO. I’ve learned a tremendous amount about AI and machine learning the past two-plus years and am very enthusiastic about how this field can continue to expand and drive improved outcomes for patients across disease areas.

Although my career path has been non-traditional for a pathologist, I’ve found it to be the perfect mix of medicine, science, business, and innovation. I enjoy sharing my career journey with young pathologists in training, so they are at least aware of the healthcare industry option given that it’s often not presented to them in medical school or residency.

What do you think makes your company stand out? Can you share a story?

What makes PathAI stand out is the passion and commitment all of our employees bring to our mission to improve patient outcomes with AI-powered pathology. I truly believe that being pathologist-led and having a relentless and data-driven focus on solving real issues in medicine and pathology are the main reasons we’re the recognized leader in the field. PathAI is uniquely positioned in the pathology and biopharma industry, combining AI-powered pathology solutions with end-to-end pathology and histology services. The strategy is to not only solve the ‘here and now’ challenges in pathology today but also to work with biopharma on the next generation of precision medicine drugs and diagnostics.

As a result of this strategy, we have solutions that pathologists can use today to improve diagnosis and laboratory operations and workflow while we also work with biopharma partners to execute clinical trials where pathology-based endpoints, biomarker classification, and/or superior histology quality are critical to successfully gauging therapeutic efficacy, accelerating new drug development for complex diseases.

But, for me, it all comes back to the outstanding level of talent we have across the organization. From the world-class machine learning team, who build the most robust algorithms for use in pathology and drug development, to our outstanding Memphis-based PathAI Diagnostics laboratory team who provide industry leading pathology services to our clinical and biopharma clients, to our dedicated contributor network of over 550 board-certified pathologists who provide critical annotation data to power our AI algorithms, we are truly privileged to have such a unique combination of outstanding talent to power our company and pursue our patient-focused mission.

You are a successful business leader. Which three character traits do you think were most instrumental to your success? Can you please share a story or example for each?

1. Authenticity. Leadership is about motivating and inspiring people and teams around common goals. I feel very strongly that this is best done in a way that’s authentic to you as a person, because people can immediately tell if you’re trying to be someone that you’re not. I’ve developed my leadership principles through incorporation of attributes from other leaders who inspire me, but in a way that fits my personality versus trying to directly copy or mimic my mentors.
2. Taking a genuine interest in people. Whether it’s colleagues, team members, customers, or collaborators, I’ve found that building strong and long-lasting relationships starts with taking a sincere interest in them as people. For example, I typically start conversations by asking about common interests, people, family, etc. before getting into the project or business topics. As a hard-core foodie, one of my favorite ice-breaker topics is to ask about people’s favorite food experiences!
3. Developing people. My favorite aspect of leadership is the opportunity and privilege all leaders have to help people develop themselves and their careers toward aspirational life goals. My approach is to ask about people’s career interests outside of their current role and then to assist them in the pursuit of new opportunities in those areas. One good example is a pathologist on my team who responded to this question by telling me that she had an interest in exploring international commercial roles, which was very different than the microscope-based jobs she had her entire career. Six months later, I helped connect her with our general manager in Spain, who hired her into a commercial medical affairs role in Barcelona. Two years after that she became the site leader for a different segment of our business and has gone on to have several other leadership and business roles in different companies. I continue to enjoy this aspect of leadership the most because it can be very surprising to learn about hidden talents that people have and extremely gratifying to see people blossom in previously unexpected ways in a new role and career path.

Let’s now move to the main point of our discussion about AI. Can you explain how AI is disrupting your industry? Is this disruption hurting or helping your bottom line?

The disruptive aspect of AI lies in the ability of this technology to process large amounts of image and language data at scale in ways that identify novel disease relationships and correlations that are beyond human capabilities. We’re just beginning to understand how best to apply AI to healthcare, but in my field of pathology, the initial applications focus on improving pathologists’ ability to make accurate and reproducible diagnoses. Examples include AI highlighting rare cells or areas of tumor that a pathologist may overlook and AI-enabled quantification of biomarkers that would manually require tedious counting and subjective estimation. Even more profound disruption is coming in the form of non-hypothesis driven AI methods like Graph Neural Networks (GNNs), which can be used to reveal complex cellular and tissue spatial relationships that correlate to important patient endpoints like survival and drug response but are virtually imperceptible by human eyes.

Generative AI and Large Language Models (LLMs) have the potential to help pathologists generate comprehensive pathology reports that are more useful to both clinicians and patients.

Each patient has a mountain of data about him or herself that must be considered in order to make the best diagnostic and treatment decisions. This includes imaging data like CTs and MRIs, anatomic and molecular pathology data, laboratory results, and genomic and genetic data. In addition, new medical guidelines, FDA approvals, medical literature and clinical trial data are being generated daily. It is information overload for pathologists and clinicians because it’s literally impossible to evaluate all of this data in the context of individual patient decisions, with the wrong decisions having the potential to dramatically impact a patient’s diagnosis, treatment, and ultimately their health outcomes. Generative AI and LLMs can help with the information overload problem because they are capable of ingesting all of this data in real-time and presenting suggestions to pathologists and oncologists in the form of pathology reports, microscopic descriptions, and recommendations for additional diagnostic tests, FDA approved therapies, and clinical trials. We’re still in the early phases of exploring these solutions, but even the initial efforts and output have been groundbreaking and indicative of a true revolution in the near future.

AI also has the capability to transform operational and workflow processes in pathology that have essentially remained static for decades. Here, the disruption comes in the form of AI pre-processing of digital pathology slides before they are even seen by the pathologist. This unlocks novel workflows that have the potential to improve both efficiency and quality. Examples include AI-enabled sorting of patient cases so that the cancer-containing samples are brought to the top of a pathologist’s worklist and related AI algorithms that trigger the ordering of appropriate molecular tests before the pathologist’s review, both of which will reduce the overall diagnostic turnaround time, enabling the clinician to notify and treat the patient sooner. AI pathology algorithms can also function in a quality control (QC) capacity, automatically flagging cases that need to be re-processed, freeing up technician and pathologist time that can be applied to core diagnostic tasks. In our Memphis lab, digital pathology has already proven to be cost-effective, with the ROI coming in the form of courier and shipping cost avoidance.

Despite the disruptive potential of these AI pathology tools, the vision of AI in pathology is not to replace the pathologist, but to enhance their work by making them more accurate, precise, and efficient, ultimately enabling them to provide the most relevant treatment-guiding information to clinicians and patients to drive the best patient outcomes.

Which specific AI technology has had the most significant impact on your industry?

AI technology overall is rapidly evolving and becoming more powerful, and this is certainly true as it applies to the field of pathology. Digital pathology refers to the process of digitizing glass slides to generate whole slide images (WSIs) that can be viewed by pathologists for diagnostic purposes. While WSIs alone enable new pathology use cases like remote diagnosis (e.g. pathologists making diagnoses from home during the pandemic) and remote consultations, AI can be applied to WSIs to unlock even more powerful capabilities and insights that have direct benefits for treating physicians and patients.

The most commonly used AI technology in pathology today is called convolutional neural networks (CNNs). CNNs are algorithms that can be trained to recognize cellular and/or tissue features that have been identified by human pathologists — for example, tumor cells or invasive tumor areas in a breast biopsy sample. Once the training is completed, these AI models can help pathologists identify these cells, tissue areas, or other structures in new, independent samples. The key advantage of this technology lies in its scalability — a human pathologist could count every single tumor cell on a slide, but since a typical slide contains hundreds of thousands of cells, it would take hours or even days. In contrast, AI models can do this in seconds or minutes, making whole slide cell characterization a reality. There are challenges of course. For example, the training could require hundreds of thousands of data points before it achieves acceptable performance. And it’s important to note that these AI models are limited to the features on which they were trained (supervised learning) and therefore will not discover any new correlations.

Another type of AI technology is referred to as end-to-end (E2E), meaning that instead of training the model to predict the presence of a specific feature like a tumor cell, the model is presented with labeled groups of cases and designed to discover any features that separate the groups (weakly supervised learning). For example, let’s say a cancer drug clinical trial has a group of subjects who respond to a treatment (responders) and a group who does not (non-responders). We want to discover a biomarker that identifies future patients who will respond. Assuming biopsy slides from patients in both groups are available in adequate numbers, the E2E AI technology may discover features in those biopsy slides that separate patients into responder and non-responder groups. Once validated, this could be the basis of a companion diagnostic that predicts which future patients will derive benefit from the drug. One form of this technology that’s been applied to pathology slides is called multiple instance learning (MIL). The advantage versus CNN technology is the potential to discover new predictive biomarkers that didn’t previously exist. One downside is that these methods tend to be more “black-box,” meaning that a human pathologist may not be able to visualize the cellular or histologic features that are driving the prediction. PathAI has discovered and patented a novel form of MIL called additive MIL (aMIL) that brings more “explainability” to this field.

Can you share a pivotal moment when you recognized the profound impact AI would have on your sector?

This moment for me came years ago when the first data was published on so-called molecular-prediction H&E algorithms. H&E, or hematoxylin & eosin, is a basic dye-based stain that all pathologists use as the basis for the initial diagnosis of disease (e.g. colon cancer). Historically, H&E slides are only used for primary diagnosis, with advanced diagnostic features requiring more sophisticated technologies like immunohistochemistry (IHC) or molecular techniques like next-generation sequencing (NGS). In the era of precision medicine, these advanced biomarkers have become critical for complete patient and tumor profiling to determine prognostic risk categories and for identification of appropriate therapies (e.g. companion diagnostics).

About five years ago, data started to emerge that some of these advanced molecular characteristics, such as gene mutations, could be predicted from the H&E slides with E2E AI methods (see discussion above). When I first heard about this data, I completely dismissed it because the pathologist and scientist in me couldn’t understand how a simple H&E slide could contain information relating to complex genomic changes. Five years later, enough data has accumulated not only to convince me that this is possible, but also to make me believe that this approach will become a standard part of the diagnostic workflow. These methods are clearly able to identify complex tissue and cellular relationships and patterns that are not accessible by human pathologists. However, since the performance is not fully equivalent to the original molecular methods, these algorithms likely will not replace them, but rather serve in a screening or triage capacity. Despite this, the benefits are profound and globally impactful because molecular biomarker technologies like NGS are time consuming, expensive, and not widely accessible, the latter especially true in developing countries.

H&E-based molecular prediction approaches have the real potential to democratize molecular diagnostics and dramatically improve access to cutting edge precision diagnostics and therapies across the world.

How are you preparing your workforce for the integration of AI, and what skills do you believe will be most valuable in an AI-enhanced future?

It is critical for pathologists across all practice environments to learn about AI, how it can benefit their practice today, and also how to best utilize the technology to drive additional value and growth in the future. When I’m giving presentations to pathologists, I tell them that they should be as familiar with AI pathology methodologies like convolutional neural networks and multiple instance learning (see discussion above) as they are with existing lab technologies like immunohistochemistry, ELISA, PCR and NGS. AI should be part of the continued professional development of a range of healthcare professionals, but I see it as imperative for pathologists given the potential across so many applications and use cases to improve diagnostic accuracy and reproducibility, and also to drive workflow and efficiency. To encourage advancement in the field, I am working with a talented and passionate group of digital pathology experts as part of the College of American Pathologists (CAP) Digital & Computational Pathology Committee on our charge to advance the adoption of digital pathology and to serve as a respected resource for information and education for pathologists, patients and the public.

What are the biggest challenges in upskilling your workforce for an AI-centric future?

The primary challenge is to overcome the mindset that the way we’ve traditionally practiced pathology in the past is just fine and doesn’t need to change. Recently, AI has received a lot of attention and, at first, I think there was fear that it could replace what pathologists do. As time has passed and the technology and specific applications have evolved, that fear has been replaced by excitement. Pathologists today realize they are being asked to do more with less and that AI can truly be their assistant to take on tedious tasks they don’t like doing anyway. A good example is PD-L1 biomarker scoring, a quantitative task that pathologists don’t enjoy — I’ve asked hundreds of pathologists if they enjoy this task, and have yet to find one who does!There has been more of a shift to embrace AI, learn more about it, and incorporate it into the daily pathology workflows in ways that drives tangible value. Education and getting to the details quickly are key to ensure integration is done well and always with the pathologist’s and laboratory needs at the forefront.

What ethical considerations does AI introduce into your industry, and how are you tackling these concerns?

As AI pathology continues to evolve as a field, ethical safeguards both at the level of industry research and development (R&D) as well as in the practice of pathology are essential to ensure ongoing trust in the field and to protect sensitive patient data and confidentiality. First, companies and organizations using and building AI for pathology need to prioritize transparency into how AI systems are developed and operate. End users need to be fully aware of the quantity and nature of the data used to create and train AI algorithms but also aware of key attributes of these algorithms such as the specific AI methodology, whether it is static or dynamically learning, and detailed performance characteristics based on relevant validation data. Another related consideration is the diversity of data that is used to train the models to avoid bias and ensure that AI models will generalize to all intended patient populations. Other ethical considerations extend to the general practice of pathology and involve the appropriate validation of AI-based diagnostic tools to ensure that they are not only safe and efficacious for their intended purpose, but also contain the necessary safeguards (e.g. data access rights) to protect patient confidentiality and protected health information (PHI).

What are your “Five Things You Need To Do, If AI Is Disrupting Your Industry”?

1. Get started– implement digital and AI pathology in your practice environment, even if that initially begins as a pilot project on a small scale. Show value in selected areas of need and expand from there.

2. Professional development — pathologists need to learn about AI methods and be fluent enough in the language of AI to effectively select the best solutions to drive growth and value in their specific practice setting.

3. Network differently — attend AI or digital pathology conferences not only to learn from the content (e.g. lectures, posters) but also to network with AI early adopters and experts.

4. Generate evidence — perform AI and digital pathology studies to demonstrate their value and ROI to internal stakeholders and decision makers like the hospital and health network CEO, CIO, CFO, etc. This evidence will also be important to drive reimbursement for AI applications in the future.

5. Share best practices — publish articles, present posters, and give presentations on your own experiences with AI pathology to share with others in the field.

What are the most common misconceptions about AI within your industry, and how do you address them?

There are several misconceptions about AI pathology:

1. That AI will devalue the role of the pathologist and/or replace pathologists. Thankfully, this is far less frequent today than in the past. Since pathologists are being asked to do more and more with fewer resources, they now view AI pathology as a valuable assistant technology that will pre-process information, perform tedious tasks (e.g. counting cells) and help solve the information overload problem by integrating vast amounts of healthcare data into digestible forms.
2. That all AI pathology is like ChatGPT: As described above, there are multiple AI methods being used in pathology including CNN, E2E/MIL, as well as Generative AI and LLMs. Each of them has specific strengths and limitations that make them more or less suitable for specific pathology use cases and applications. Training and education efforts as discussed above will address this challenge over time.
3. That all AI pathology algorithms are “learning” in real-time. In reality, most pathology AI applications today are static, meaning that although they have been dynamically trained on large amounts of input data, they are typically locked and not learning from pathologist input in the field. Transparent communication of AI product profiles combined with education will correct this misconception.

Can you please give us your favorite “Life Lesson Quote”? Do you have a story about how that was relevant in your life?

My late father advised me at an early age to “follow your bliss” and make decisions based on what I was passionate about, regardless of titles, money, or other external factors. As an OB/GYN who brought over ten thousand lives into the world and helped hundreds of women with infertility issues have children, my father taught me this life lesson through example. He truly never worked a day in his life and thanks to him, I feel the same way about my life and career in pathology and the healthcare industry.

Off-topic, but I’m curious. As someone steering the ship, what thoughts or concerns often keep you awake at night? How do those thoughts influence your daily decision-making process?

Usually this question is meant to highlight issues or problems, but what actually keeps me up at night is the opposite — that people won’t embrace AI technology enough, which in my field may result in the delay of the benefits to pathologists, clinicians, and patients. Digital and AI pathology is at an inflection point, where the field is moving rapidly, but we collectively still have a lot of work to do to get to the point where most pathologists and the patients they serve have access to this technology. The whole field is working actively on the solutions but I’m impatient; I’m doing everything I can to help educate pathologists on AI technology and have also created a program for the next-generation of pathologists, pathology residents in-training, to learn about AI methods in the setting of our R&D program at PathAI.

You are a person of great influence. If you could start a movement that would bring the most amount of good to the most amount of people, what would that be? You never know what your idea can trigger. 😊

The movement would be to ensure that all patients around the world have easy access to their pathology reports and that they actually understand the report content before they meet with their treating physician.

Patients often reach out to me seeking assistance with their pathology diagnosis and cancer treatment plan. The first question I always ask is: “Do you have your pathology report?” I’m shocked by how frequently patients do not have access, which is the primary diagnostic document summarizing their disease, including all of the accessory data (e.g. molecular sequencing) that was performed as part of the diagnostic work-up. Patients typically don’t even know that such a report exists — they hear about their diagnosis verbally from their physician, but often this is in very general terms and leaves out valuable information that could be critical for downstream treatment decision making.

The second part of the problem is understanding the pathology report. Pathology reports are not written in a style and with language easily understandable by patients. Even physicians can have trouble fully understanding all the information in a pathology report. For example, a recent study (Gibson et al. Arch Pathol Lab Med 2022) found that only 52% of staff physicians reported being very comfortable reading a pathology report. AI technology, including large language models could significantly help here, in essence, creating a way to translate pathology reports, not only standardizing terminology, but going one step further to extract the key information and present it to treating physicians and patients in a format that enables optimal, rapid, and clear treatment decisions.

How can our readers further follow you online?

Readers can connect with me on LinkedIn and read about PathAI’s most recent work at pathai.com.

Thank you for the time you spent sharing these fantastic insights. We wish you only continued success in your great work!

About the Interviewer: Cynthia Corsetti is an esteemed executive coach with over two decades in corporate leadership and 11 years in executive coaching. Author of the upcoming book, “Dark Drivers,” she guides high-performing professionals and Fortune 500 firms to recognize and manage underlying influences affecting their leadership. Beyond individual coaching, Cynthia offers a 6-month executive transition program and partners with organizations to nurture the next wave of leadership excellence.