Joseph Nano Uncategorized

AI in Cancer Pathology: Big Promise, Slow Rollout

A new expert review says it’s time for a “reality check”—and a clearer path from lab to clinic.

Artificial intelligence is rapidly getting better at reading digitized microscope slides, spotting cancer, and even predicting which patients might benefit from specific therapies. But despite recent breakthroughs, very few AI tools are actually approved and used in day‑to‑day pathology practice, according to a 2019–2024 field review in Nature Reviews Clinical Oncology.  

What’s new

  • Smarter models, broader skills. Pathology AI has leapt from classic CNNs to “foundation models” and multimodal systems that combine slide images with clinical data and genomics. These systems have posted strong results on tasks like tumor detection, immune‑cell mapping, and segmentation.  
  • Regulators are moving—but slowly for pathology. By late 2024, the FDA had authorized >1,000 AI/ML medical devices overall. Yet radiology accounts for 76% of them, while pathology has just three (<0.5%). Median review times are longer for pathology than radiology, reflecting tougher integration and evidence hurdles. Figure 1 on page 285 shows the gap at a glance.  
  • First wave of real‑world deployments. Approved tools cover tasks like prostate cancer detection and immunohistochemistry quantification; early data suggest these systems can speed reads, standardize scoring, and assist tumor boards.  

Why adoption still lags

  • Evidence bar not yet met. Most products stop at “level III” evidence (unblinded external validation). The authors lay out a four‑step roadmap culminating in AI‑informed randomized trials; no digital pathology model has reached level I evidence yet. See the validation pathway diagram on page 286.  
  • Infrastructure is heavy. Hospitals need scanners, storage, fast networks, and workflow integration—costs that have slowed global roll‑out, especially outside large centers. Still, case studies report shorter turnaround times and notable cost savings once digital systems are in place.  
  • Payment and policy are in flux. In the U.S., 30 new Category III CPT add‑on codes (2024) now track slide digitization and could mature into standard reimbursed codes; Europe is advancing its AI Act and IVDR pathways. Debate continues over laboratory‑developed tests (LDTs): one AI‑enabled prostate test already carries a PLA code (0376U) and appears in 2024 NCCN guidelines, but FDA oversight of LDTs is tightening.  

What this means for patients

If done right, AI in pathology could reduce diagnostic delays, improve consistency, and better match treatments to individuals—especially as models learn from images plus clinical and genetic data. But the review urges cautious optimism: demand stronger trials, clear reporting, and fair reimbursement so benefits reach routine cancer care safely and equitably.  

Source: Aggarwal A, Bharadwaj S, Corredor G, et al. “Artificial intelligence in digital pathology — time for a reality check.” Nature Reviews Clinical Oncology. April 2025.  

Aggarwal, A., Bharadwaj, S., Corredor, G., Pathak, T., Badve, S., & Madabhushi, A. (2025). Artificial intelligence in digital pathology – time for a reality check. Nature reviews. Clinical oncology, 22(4), 283–291. https://doi.org/10.1038/s41571-025-00991-6

Editor’s note: This article is for information only and is not a substitute for professional medical advice.