AI in Pathology Is Only as Good as the Data It Can Trust

Artificial intelligence (AI) has recently been discussed as the future of pathology during these years. Whether the conversation involves digital pathology, cancer detection, workflow optimization, or case prioritization, AI is increasingly being presented as a major part of the future laboratory environment.

ASCO 2026

This trend was particularly noticeable during the current 2026 ASCO meeting. Walking through the exhibition floor,  it became clear that AI is no longer a niche topic in pathology. New companies are entering the field at a rapid pace, while many established diagnostics companies are investing heavily in AI-driven solutions. Some systems focused on tumor detection, others on case triaging and many other applications. Regardless of the specific technology, the message was remarkably consistent: AI is expected to help pathology laboratories work faster and supporting pathologists in making more informed decisions in the near future.

The excitement is justified. AI systems are becoming increasingly capable of analyzing large numbers of digital slides, identifying findings that deserve closer review. For laboratories facing growing workloads and staffing challenges, these tools have the potential to significantly improve efficiency. Perhaps more importantly, AI does not get tired. While a pathologist reviewing hundreds of slides throughout a busy week may naturally experience fatigue, AI systems can apply the same criteria repeatedly and consistently across thousands of images and never feel tired. This consistency is one of the biggest reasons many laboratories are exploring how AI can become a valuable assistant within the diagnostic process.

AI Pathology

However, AI also has limitations. Despite the impressive advances in image analysis, AI does not truly understand the specimen it is analyzing. It cannot independently verify clinical context or diagnostic accuracy without human oversight. Most importantly, it cannot determine whether a slide has been linked to the correct patient. Like any analytical tool, AI can only work with the information it receives. In other words, AI is only as reliable as the data it receives.

This raises a question that is discussed far less frequently than AI algorithms themselves: How does AI know which patient it is looking at?

Before a slide can be analyzed by either pathologist or AI platform, it must first be correctly identified. The specimen must be connected to the correct patient or correct case record. If this connection fails, the sophistication of pathologist or AI becomes largely irrelevant. No matter how accurate the analysis may be, if the specimen being analyzed is linked to the wrong patient, the entire result becomes meaningless. In many ways, AI is only as reliable as the specimen tracking systems that support it.

Well printed label

This is where standardized identification becomes increasingly important. As pathology laboratories continue moving toward digital workflows, barcode-based identification systems allow specimens to be tracked throughout the entire process such as from grossing and embedding to sectioning, staining, scanning, and long-term archiving. A barcode can be scanned instantly, linked directly to the LIS, and retrieved automatically whenever needed. The physical specimen and its digital record remain connected throughout the workflow.

Handwrite label

On the other hand, handwritten labels present a very different challenge. Experienced laboratory staff can often interpret handwriting with little difficulty. AI systems cannot. Handwriting is difficult to standardize and integrate into large scale digital workflows. What works reasonably well for human interpretation often becomes a barrier for machine-based processing. As laboratories begin managing thousands or even millions of digital pathology images, manual verification becomes increasingly difficult, while automated tracking becomes increasingly necessary.

For this reason, the future of AI in pathology may not depend solely on more powerful algorithms. It may also depend on the infrastructure supporting those algorithms. Slide scanners, LIS, barcode readers, tracking systems, and standardized labeling practices rarely receive the same attention as AI software. Yet they provide the foundation that allows AI to function reliably in the first place.

AI will likely become an extremely capable assistant for pathologists in the years ahead. It may help identify patterns faster and improve efficiency. But before AI can analyze a slide, it first needs to know exactly whose slide it is looking at.

And that begins with accurate, standardized specimen identification long before the algorithm ever sees the image