On the Human Health, Impact and Technology webinar series on April 1st 2025, Dr Andy Beck MD PhD, Chief Executive Officer and Co-Founder at (opens in a new window)PathAI spoke to host Professor Patricia Maguire about "AI-powered Diagnostics".
From Pathology to AI
Andy trained in pathology, and he’s always been fascinated by how it sits at the crossroads of science and medicine. In pathology, tissues removed from a patient are sent to the lab, and it’s the pathologist’s job to analyze those samples and provide a diagnosis. What intrigued Andy most, particularly as a student, was the process of examining microscopic images. The ability to study these images, decode them, and translate that into diagnoses to guide therapy was always what captivated him. Over time, his focus expanded from simply building a vast mental library of images to exploring how we could enhance human intelligence with artificial intelligence. Andy became interested in the hypothesis that a pathologist, supported by AI, could be more accurate, reproducible, and efficient than one working alone. This led him to pursue a PhD in machine learning, a field he knew little about after his pathology training. Following his PhD, Andy spent five years in academia researching this area before founding PathAI in 2016. The company is dedicated to improving patient outcomes through AI-powered pathology.
How PathAI is Enhancing Pathology
A key insight into understanding pathology is that much of the work is still highly manual. Experts are trained by previous generations of experts to analyze a large number of complex images and make accurate diagnoses. While this process works well for many cases, there is often disagreement between pathologists on certain cases. Additionally, it’s challenging for humans to derive novel predictors for which treatments will work best based on visual data alone. As the treatment landscape continues to evolve, it becomes increasingly important to extract as much data from these images as possible.
PathAI’s technology addresses this by training machine learning models on tens of millions of expert-annotated images. Over the past decade, machine learning has become highly effective at learning from these examples, identifying patterns, and predicting those patterns on new, unseen images. For research purposes, they use fully automated workflows where the input is images and the output is computer-derived features. However, for clinical applications, they see AI as a tool to augment pathologists. They've developed a platform that allows pathologists to use a computer monitor rather than a microscope to analyse images. For certain tasks, pathologists are supported by AI models.
Expanding PathAI’s Impact in Diagnostic Labs
In 2016, PathAI started as a purely software company, initially focused on serving drug development labs. After recognizing the potential to impact patient care, the company shifted its focus to diagnostic labs. In 2021, after assessing the lab landscape—particularly in the U.S.—they discovered that over 95% of labs were still using traditional microscopes and glass slides, rather than digital technology. PathAI began helping labs transition to digital pathology technology. Unable to find enough partners ready to move quickly, PathAI purchased their own lab in 2021 and spent the next three years developing a platform to support diagnostic pathologists. However, they realized their global vision was better achieved by providing technology to labs rather than running their own.
In 2024, PathAI had the opportunity to partner with Quest Diagnostics, the largest pathology lab in the U.S. This partnership not only involved selling their laboratory to Quest, but also made them PathAI's first major diagnostic lab partner. Quest has committed to large scale implementation of digital pathology and, over time, expect to increasingly incorporate AI to support their pathology practice.
Key Diseases PathAI Digital Tools Help Detect
PathAI focuses on areas where pathology slides play a critical role in diagnosing diseases. While they don’t work in areas like psychiatric, neurologic, or most cardiovascular diseases—since these conditions don’t typically involve tissue biopsies—they are deeply involved in a wide range of other conditions. These include skin biopsies for both inflammatory diseases and cancerous conditions, such as melanoma. PathAI's work spans across nearly every solid tumor, with common tissue specimens including skin, breast, lung, prostate, and colon. Major disease areas of focus include oncology, covering both early and advanced stages of cancer, as well as inflammatory bowel disease, which involves biopsies from the GI tract, including the mouth, oesophagus, stomach, colon, and rectum. Liver disease, particularly metabolic-associated steatohepatitis (MASH), is another key area, as it’s a prevalent liver disorder where pathology plays an essential role. Essentially, wherever tissue biopsies are required for diagnosis, PathAI has developed systems to support pathologists.
PathAI’s Impact on Research and Diagnostic Accuracy
PathAI has made a significant impact in both research and diagnostics by leveraging AI to analyze hundreds of thousands of cells per image—something that can't be done manually. By extracting quantitative features from these images, PathAI helps identify new predictors of drug response, a key area in drug development. On the diagnostic side, about 90% of pathology is still done using microscopes and glass slides. PathAI’s platform revolutionizes this by enabling digital viewing and collaboration, allowing labs to digitize glass slides and share them with pathologists worldwide. This enhances collaboration, speeds up case review, and improves workflow efficiency. Many labs adopt digital pathology first for these workflow and collaboration benefits.
Advancing Precision Medicine
One of the most exciting aspects of this technology is its broad applicability across various diagnostic fields. Unlike many diagnostic tools that are highly specific, PathAI’s digital pathology viewer and collaboration tools add value to every type of slide, as they can be digitized and viewed regardless of the diagnosis. While there are many areas of focus, one that particularly excites us is the ability to predict alterations from just a standard H&E image—something that is invisible to the naked eye. Recently, we published a study showing that we can predict a rare type of aberrant protein expression from an H&E image alone. This could help identify patients who would benefit from a specific test they might not have otherwise received. Looking ahead, we envision a future where patients worldwide are screened by AI systems that flag potential rare alterations, prompting confirmatory tests and, potentially, targeted treatments that could cure specific subtypes of diseases. While oncology is the area where this technology is most advanced, we expect to see similar precision medicine approaches in other diseases, including inflammatory bowel disease and liver disease, in the coming decade.
Fair and Ethical AI: Serving Diverse Patient Populations
Ensuring fair and ethical AI, especially when serving diverse patient populations, is crucial. One key strategy is using a wide range of diverse samples during product development to train AI systems, ensuring the model can generalize to various populations. Even more important is validating the system on diverse populations to ensure it works effectively for everyone. Beyond development, equitable distribution of these technologies is essential. The major upfront investment in digital pathology, such as full-slide imaging scanners, can be a barrier. However, once the infrastructure is in place, AI systems can be deployed at low cost, democratizing expert knowledge. It’s far cheaper to apply an AI system to thousands of slides globally than to rely on human experts to review each one. The challenge lies in ensuring fair access to this technology worldwide, but once that’s achieved, AI can be a powerful tool for democratizing expertise across the globe.
Shared Responsibility in AI Decision-Making in Healthcare
The responsibility for decision-making in AI-driven healthcare tools lies with both the manufacturers and the physicians. Manufacturers, like us, are responsible for validating the tools, working with regulators when necessary, and ensuring these tools are safe, effective, and transparent. It’s crucial that these tools are developed by experts to meet high standards. On the other hand, physicians who use these tools also bear responsibility. They must ensure the tools are used correctly, following standard care practices, and that the systems are deployed as they were validated. Ultimately, it’s a shared responsibility—both sides must work together to ensure AI systems are used effectively and safely.
AI Literacy and Clinician Training in Healthcare
Residency programs are a key opportunity to integrate AI training for future clinicians. Both current pathologists and the younger generation are well-equipped to adopt new technologies. However, device manufacturers must ensure these tools are intuitive and have smooth user experiences. Much like the viral success of ChatGPT, which combined sophisticated AI with a simple, effective interface, AI healthcare tools should aim for similar ease of use while handling complex processes behind the scenes. Additionally, it’s crucial for clinicians to be able to evaluate and trust the tools they use. This responsibility lies not only with physicians but also with regulators and manufacturers to ensure the tools available are reliable and effective.
What's Next for AI-Powered Pathology?
The future of AI-powered pathology is poised for significant transformation. Despite its potential, digital pathology is still in its early stages. Many pathology labs continue to rely on traditional glass slides and microscopes, and getting a second opinion often requires physically shipping slides across the world, which can take days or even weeks. The first major shift will be the widespread adoption of digital pathology, allowing patients to have control over their pathology images, just as they do with personal photos, enabling them to share their diagnostic data easily and securely. Alongside this, AI will play an increasingly important role in enhancing pathologists' performance. In just a few years, we may look back at the days when pathologists examined images without machine learning support as a more primitive approach. The technology to improve pathology is already here—it’s now a matter of refining the products and ensuring that the technology is usable and accessible, with a sustainable model for widespread distribution.