Radiology gets most of the attention when healthcare AI is discussed. But pathology—the diagnosis of disease through examination of tissues, cells, and body fluids—has equally significant AI potential.

The opportunity is clear. The adoption has been slower. Here’s what’s happening.

Why Pathology AI Matters

Pathology underpins so much of medicine. Cancer diagnosis, infection identification, disease monitoring—these depend on pathologists examining specimens and making judgments.

Australia faces pathology workforce challenges. Pathologist numbers aren’t keeping pace with demand. Subspecialty expertise is concentrated in major centres. Regional areas struggle with access.

AI could help by:

• Improving pathologist productivity (more specimens reviewed per hour)
• Reducing subjective variation (more consistent diagnoses)
• Enabling remote expertise (AI-assisted review without physical specimen transport)
• Supporting quality assurance (automated checking of diagnoses)

The potential is real. So why is adoption lagging radiology?

The Technical Challenges

Pathology AI is technically harder than radiology AI for several reasons:

Image scale. A single histopathology slide can generate a digital image of 100,000 x 100,000 pixels—ten billion pixels total. Processing images this large requires significant computational resources.

Scanning infrastructure. Digital pathology requires slides to be scanned into digital images. This scanning infrastructure is expensive and not universal. Many pathology labs still work primarily with glass slides under microscopes.

Annotation complexity. Training AI requires pathologists to annotate images—marking regions of interest, identifying cell types, indicating abnormalities. This annotation is time-consuming and requires subspecialty expertise.

Workflow integration. Pathology workflows are complex and variable. Integrating AI into existing laboratory information systems is challenging.

These aren’t insurmountable, but they explain why pathology AI development has been slower than radiology.

What’s Actually Working

Despite challenges, some pathology AI applications are delivering value:

HER2 scoring in breast cancer. AI for assessing HER2 protein expression in breast cancer specimens has strong evidence. HER2 status determines treatment eligibility, so accurate, consistent scoring matters. AI can match expert pathologist performance and reduce inter-observer variability.

Ki-67 quantification. Ki-67 is a proliferation marker used in cancer assessment. Manual counting is tedious and variable. AI quantification is faster and more reproducible.

Prostate cancer grading. AI for Gleason scoring of prostate cancer has advanced significantly. Several products have regulatory approval and clinical validation.

Metastasis detection. AI for identifying cancer metastases in lymph node specimens can serve as a “second reader,” catching findings that might be missed.

These applications share characteristics: well-defined tasks, quantitative outputs, strong clinical evidence. They’re not “AI reads pathology slides”—they’re AI performing specific tasks within the broader pathology workflow.

What’s Struggling

Broader pathology AI adoption faces challenges:

General diagnostic AI. AI that handles the full range of pathology diagnosis—like a general pathologist does—doesn’t exist. Current AI is narrow, focused on specific tasks.

Rare conditions. AI trained on common diseases performs poorly on rare conditions. Pathology includes many rare diseases where AI training data is limited.

Quality variability. Specimen preparation, staining, and scanning quality vary between labs. AI trained on high-quality images may perform worse on routine clinical images.

Pathologist acceptance. Some pathologists see AI as a threat. Others doubt AI can handle the complexity of diagnostic pathology. Cultural resistance is real.

The Adoption Gap

Radiology AI has seen broader adoption than pathology AI. Several factors explain this:

Digital workflow maturity. Radiology went digital years ago. PACS systems, digital imaging, electronic reporting—the infrastructure for AI integration exists. Digital pathology is less mature.

Image complexity. Radiology images, while large, are more standardised than pathology images. Radiology AI integration is technically simpler.

Business model. Radiology AI can often demonstrate productivity gains more clearly. Pathology productivity measurement is more complex.

Early vendor investment. More AI vendors targeted radiology early, creating more product choice and competition. Pathology AI had fewer options.

This gap is closing. Digital pathology adoption is accelerating. Pathology AI products are proliferating. But pathology is two to three years behind radiology in AI adoption maturity.

Australian Context

Australia’s pathology sector has specific characteristics that affect AI adoption:

Market concentration. A few large pathology providers dominate the private market. These organisations have the scale and resources to invest in AI, and some are doing so.

Public pathology variation. Public hospital pathology services vary significantly in digitalisation and AI readiness. Some are advanced; others are significantly behind.

Geographic challenges. Regional pathology services face technology and workforce challenges that could make AI particularly valuable—but also make implementation harder.

Regulatory pathway. TGA provides clear pathways for pathology AI as medical devices. This isn’t a barrier, though navigation requires effort.

My Assessment

Pathology AI is at an inflection point. The technology is maturing. Digital pathology infrastructure is expanding. Products are accumulating evidence.

For healthcare organisations considering pathology AI:

Start with specific, validated applications. HER2, Ki-67, prostate grading—these have the strongest evidence and clearest value propositions.

Assess digital pathology readiness. If you’re not digitised, that’s the first investment. AI requires digital images.

Engage pathologists as partners. Pathologist resistance can derail adoption. Early engagement and clinical leadership are essential.

Plan for infrastructure. Storage, computing, and integration requirements are significant. Understand the full cost picture.

Watch for emerging applications. The field is moving quickly. Applications that aren’t ready today might be ready in 12-18 months.

Pathology AI won’t transform the field overnight. But the trajectory is clear: AI will become an integral part of diagnostic pathology. The question is when, not if.

---

Dr. Rebecca Liu is a health informatics specialist and former Chief Clinical Information Officer. She advises healthcare organisations on clinical AI strategy and implementation.