You were among the earliest champions of digital and computational pathology. What pivotal moment or case convinced you that these technologies would reshape diagnostics?

A pivotal moment for me came in October 2018, when Google AI introduced a deep-learning system called Lymph Node Assistant (LYNA) for detecting metastatic breast cancer in lymph node slides. The system achieved high accuracy and was able to identify small metastatic foci that could be missed by human review. It also showed that, when used alongside pathologists, it could reduce slide review time.

Importantly, the researchers emphasized that further validation in real-world clinical settings was needed before routine use. Even so, this marked a shift in how AI was viewed in pathology. The conversation moved from “AI might help” to “AI can assist with real diagnostic tasks.”

This moment also signaled a broader change beyond the technical achievement. It demonstrated that large technology companies were investing seriously in computational pathology, which helped raise the profile of the field. It encouraged further research, collaboration, investment, and increased expectations around validation – such as testing across institutions, handling real-world variability, and evaluating how pathologists interact with AI tools.

Overall, it showed that AI could extend the capabilities of pathology, while also highlighting the need for careful implementation and validation.

How can chairs and clinical leaders ensure that computational tools are implemented in ways that enhance diagnostic accuracy and safety – not just workflow efficiency?

Implementation should start with a clear clinical need. Leaders should confirm that any tool addresses a real diagnostic challenge and aligns with patient care priorities. A structured governance framework is essential, covering areas such as tool selection, bias, validation, performance thresholds, risk assessment, and ongoing monitoring for version changes or drift.

Clinical validation is critical. Strong performance in development studies does not guarantee accuracy in real-world settings. AI tools should be validated in the local clinical environment before being used in patient care. Maintaining human oversight is also important, particularly when defining when AI can be relied upon, when it can be disregarded, and how to handle discrepancies.

Ongoing quality assurance is needed to ensure tools remain accurate and reliable over time. In addition, education and training are essential to prevent overreliance on automation. Maintaining user competency is a key part of ensuring safe and effective use of AI in clinical practice.

How do you see the role of academic leadership evolving as AI becomes more integrated into diagnostics?

Modern pathology leaders must also act as connectors between disciplines – working not only with clinicians and pathologists, but also with computer scientists, data engineers, and industry partners. This includes developing new workforce models where AI supports, rather than replaces, human expertise. Training and upskilling of both staff and trainees will be essential.

AI literacy is becoming a core requirement for leadership. Leaders need to understand both clinical and computational perspectives, while also addressing emerging challenges such as data sharing, liability, and bias. This includes establishing governance structures for validation, monitoring, and incident reporting, as well as ensuring ethical oversight.

Engagement with industry also requires careful management, including consideration of intellectual property, data use, and potential conflicts of interest.

Finally, academic leaders play an important role in shaping education. Integrating AI into medical and pathology training can help redefine the specialty and may encourage more trainees to enter the field.

How can pathologists lead in shaping responsible frameworks for safe and clinically meaningful AI integration?

Data quality is particularly important. While large datasets were once seen as essential, it is now clear that quality matters more than quantity. Pathologists can lead efforts to ensure accurate annotation, establish consensus-based “ground truth,” and include relevant metadata such as staining methods, scanner type, and tissue origin. They can also promote diverse and representative datasets to reduce bias and improve generalizability.

Another priority is ensuring that AI addresses meaningful clinical problems. Many tools have been developed for tasks that offer limited real-world benefit. Pathologists are best placed to identify where AI can add value – particularly in ways that support workflow efficiency and reduce workload, rather than adding complexity.

Pathologists also have a central role in validation, implementation, and oversight. This includes setting expectations for explainability, ensuring tools are rigorously tested, and monitoring performance over time. They can help define when human oversight is required and contribute to guideline development and policy decisions.

Finally, pathologists should help shape the narrative around AI in medicine. Framing AI as a tool that supports, rather than replaces, clinical expertise is important to ensure realistic expectations and responsible adoption.

Looking ahead, which trends in digital and computational pathology do you think will most influence diagnostic medicine in the coming decade?

One major trend is the growing use of AI-augmented diagnostics, where AI tools act as decision support for pathologists. Alongside this, multimodal data integration is becoming more important. Combining pathology images with clinical, molecular, and other data types may help detect complex disease patterns that are not easily identified through a single source.

Predictive computational tools are also emerging. These approaches aim to move pathology beyond descriptive diagnosis toward supporting treatment decisions, potentially reducing the need for additional invasive testing. In parallel, generative AI may help streamline tasks such as data analysis and reporting, and support the development of improved models through synthetic data.

Another emerging area is the development of foundation models – large, pretrained AI systems designed for general use. These models are well suited to pathology, where data are complex, variable, and often multimodal. Unlike traditional approaches that require separate models for each task, foundation models can be trained on large, diverse datasets and then adapted to multiple applications.

This shift may allow pathology AI to move from narrow, task-specific tools to more flexible systems that can be applied across different diagnostic tasks. As a result, development may become more efficient, and models may perform more consistently across varied clinical settings.

What factors will affect the integration of these tools into clinical practice?

Rather than replacing pathologists, these technologies are expected to support faster and more precise diagnoses and contribute to more personalized patient care. However, widespread adoption will depend on broader implementation of digital pathology, including whole-slide imaging and integrated platforms, often supported by cloud-based systems.

These digital foundations are essential for scaling AI in clinical practice. Regulatory frameworks and reimbursement models will also be needed to support routine use.

Newsletters

Receive the latest pathologist news, personalities, education, and career development – weekly to your inbox.

I have read and understand the Privacy Notice*

Subscribe