[Reporter's View] What is lacking in the trend of AI pathology?

Personally, I first encountered the topic of digital pathology in 2019 while entering the healthcare sector. At the time, the Korean Society of Pathologists was labeled of being a specialty avoided by medical residents, prompting intense discussions on restructuring training paradigms while simultaneously exploring ways to integrate artificial intelligence (AI).

The primary objective was to overcome the stereotype of pathology being a legacy specialty destined to fall behind in the AI era, and to establish a robust foundation for big data collection and research through virtualization. Furthermore, these discussions extended to securing health insurance reimbursement for digital pathology implementation.

Back then, the major hurdle to digital pathology adoption was hospital demand. In the absence of a dedicated reimbursement code, only a handful of large medical institutions could afford to acquire such expensive hardware and software systems. Even the major tertiary hospitals, the so-called 'Big 5' in South Korea, were highly conservative about implementation.

As the "Major AI Transition" topic spread, "AI" naturally became the first word that came to mind when discussing digital pathology. It easily conjures images of glass slides being digitized into high-resolution images, AI identifying malignant cells, and algorithms assisting pathologists in their readings. Medical AI developers frequently pitch their technological competitiveness using words such as lesion detection, biomarker analysis, and support for companion diagnostics (CDx).

However, field reports on digital pathology indicate that industry insiders still place far greater emphasis on Digital Transformation (DX) than on AI Transformation (AX). In other words, before moving to the stage of using AI for lesion detection and diagnostic support, the foundational pathology workflow must be re-engineered to fit a digital environment.

For pathology AI to function optimally, certain prerequisites must be met. ​The entire pathology lifecycle, from specimen accessioning, slide preparation, scanning, and storage to interpretation, reporting, and hospital information system (HIS) integration, must be digitally structured. Merely scanning glass slides into image files is insufficient. There must be full traceability of how a specimen was processed and how that processing correlates with specific images and diagnostic outcomes.

Unlike radiology, which underwent digital transformation relatively early, backed by PACS (Picture Archiving and Communication System), the field of pathology remains heavily dependent on analog processes. Pathology images are large, and color reproduction and standardization pose significant technical challenges. Furthermore, interoperability suffers when equipment and software architecture vary across different hospital networks. ​Compounded by high initial capital expenditures (CapEx), ongoing maintenance costs, a lack of institutional reimbursement, and friction with staff adaptation, transitioning to digital pathology is far from a straightforward decision for hospital leadership.

Skipping this foundational gap to focus prematurely on AI disrupts the industry. If AI solutions are pushed forward without sufficient digital infrastructure deployed at the clinical site, their real-world utility will inevitably be constrained. No matter how advanced an AI algorithm is, it cannot be sustained in clinical practice if the input data remains disperse, unstructured, and disconnected from the hospital’s operational workflow.

The CEO of a digital pathology company I recently interviewed similarly defined the starting point as workflow integration. It means that hospitals must first establish internal, functional baseline digital pathology capabilities before scaling up to AI-powered analytics and precision medicine.

Global players are moving in the same direction. In May of this year, Roche signed a definitive agreement to acquire PathAI, a company specializing in digital pathology and AI-powered pathology technologies. The deal is valued at up to $1.05 billion, including upfront payments and bio-buck milestones. Rather than a transaction aimed purely at short-term revenue, this acquisition serves as clear evidence that pathology data and AI-based diagnostic capabilities are increasingly viewed as indispensable infrastructure for the era of precision medicine.

Indeed, industry consensus holds that the value of pathology data will inevitably grow exponentially in the precision oncology and personalized medicine space. However, this transition will not happen automatically. Hospitals need clear financial incentives to invest in digital pathology, and pathology departments must be willing to adopt entirely new operational systems. ​Inter-system interoperability and data standardization are mandatory baselines. Furthermore, validation and accountability frameworks must be established to safely translate AI-generated insights into clinical care.

To discuss the future of digital pathology, we must focus on Digital Transformation (DX) before AI Transformation (AX). Only when the pathology workflow shifts entirely to digital, data becomes standardized, and systems connect seamlessly with broader hospital networks can AI truly become the language of clinical practice. The starting point for pathology AI is not the algorithm, but the infrastructure.