Artificial intelligence (AI) is increasingly entering the pathology lab. Research from UMC Utrecht shows that AI not only accelerates work but also improves the quality of diagnostics. Two applications illustrate how technology is changing the work of pathologists.
In the pathology lab of UMC Utrecht, clinical epidemiologist and physician-researcher Carmen van Dooijeweert and research analyst Sven van Kempen are working on how artificial intelligence can improve the work of pathologists. Their research shows that AI not only saves time but also improves the quality of diagnostics. Van Dooijeweert focuses on detecting breast cancer metastases using a smart algorithm, while Van Kempen uses AI to accurately monitor the quality of immunohistochemical stains.
Smarter search in the sentinel lymph node
In breast cancer, the sentinel lymph node is often removed after the tumor is removed. This lymph node is the first site where tumor cells end up if they metastasize. Pathologists cut the lymph node into wafer-thin sections and examine the resulting digital images for tumor cells. If they see nothing suspicious, they often perform additional staining that reveals tumor-specific proteins. "This is an important prognostic determination," explains Van Dooijeweert. "If tumor cells are present in the lymph node, it indicates the risk of further metastasis and helps determine the patient's subsequent treatment."
These stains are expensive and labor-intensive. "On average, we make five sections per lymph node, and one stain costs around 25 euros. Per patient, that easily costs well over a hundred euros. Fortunately, we don't find metastases in two-thirds of women, but that does mean many stains later." To make that process more efficient, her team investigated the use of artificial intelligence.
In the study, the team tested an algorithm from Visiopharm: the Metastasis Detection app, which scans lymph node sections and highlights suspicious areas with color codes: red for almost certain tumor, orange for suspicious, and yellow for extra attention. "The pathologist then specifically examines those areas," explains Van Dooijeweert. "This eliminates the need to examine the entire section." The result: pathologists work a third faster, don't miss any relevant metastases, and can significantly reduce the number of expensive stains. Moreover, it increases job satisfaction for pathologists. "Assessing lymph nodes is tedious work for pathologists; this is an application they're particularly happy with."
AI as a quality controller
While Van Dooijeweert focuses on efficiency, Van Kempen uses AI to monitor the quality of immunohistochemical stains. His project, which also uses an algorithm from Visiopharm: Qualitopix, measures the color intensity of so-called control sections.
"For each immunostaining test, we used old patient tissue as a control on the slides containing the patient tissue to be tested," explains Van Kempen. "But that tissue is running out and varies from case to case. That's why we now work with standardized cell lines. An AI algorithm assesses the staining strength of these controls and can detect variations or errors in the staining that are difficult to see with the naked eye. This way, we know much better when the staining quality is insufficient."
When the team tested the method, they discovered unexpected fluctuations between stainings, even on identical stainers. "We saw differences between machines and even between positions on a single stainer," says Van Kempen. "After a thorough maintenance check, the results suddenly stabilized. This showed that this algorithm is very sensitive to quality issues that we would otherwise never have discovered."
Van Dooijeweert adds: "This isn't about saving time or money, but about improving the quality of care. This algorithm helps us diagnose more accurately and prevents a patient from receiving the wrong treatment based on a subtly incorrect staining. That's a pure improvement in quality."
From potential to practice
The two studies demonstrate the broad applicability of AI in pathology. However, large-scale implementation remains a challenge. "AI is not yet reimbursed in the Netherlands through the regular DBC system," says Van Dooijeweert. "That complicates the business case for hospitals. You have to purchase software, integrate it into your systems, and often hire additional IT staff."
Still, Van Dooijeweert is optimistic about the future: "We shouldn't wait until everything is perfect or until manufacturers provide complete transparency about their training data. We simply have to start testing in practice and demonstrate that it works. That requires a pragmatic approach."
Van Kempen agrees: “We are only at the beginning. AI will soon be in almost every step of the process. But there will always be a human in the loop: the pathologist remains ultimately responsible.”
Van Dooijeweert concludes: "AI won't replace work, but enhance it. Especially with the growing number of cancer diagnoses, this is not only desirable but essential. If we want to future-proof healthcare, we can no longer do without smart support in the lab."
Want to learn more about the use of AI in the pathology lab? Sven van Kempen will be speaking at the LabAutomation event on March 10th at Congrescentrum 1931 in 's-Hertogenbosch. His presentation will address recent AI innovations at UMC Utrecht, including automated sample sectioning and automated mitosis detection.
Nederlands