AI-Powered Pathology: Insights from Dr. Aleks Żuraw
AI in pathology has evolved far beyond simple image recognition. In a recent video,Dr. Aleks Żuraw explained that today's innovations fall into three key categories that are transforming how pathologists work and how we understand disease.
🧠 1. Image Analysis
Image analysis represents the most mature category of AI in pathology, with applications spanning from basic quantification to advanced diagnostic assistance.
Quantification Tasks
AI can automatically count nuclei, measure tissue areas, and assess biomarkers like Ki-67 — removing the need for time-consuming manual counting. This not only saves hours of pathologist time but also increases accuracy and reproducibility.
Computer-Aided Diagnostics
The first FDA-cleared AI tool (Paige) demonstrated a remarkable improvement: diagnostic accuracy increased by 7 percentage points. The workflow is particularly powerful: AI highlights suspicious regions, while the pathologist confirms findings — a "human + AI" collaboration that's both faster and more accurate than either alone.
Predicting Molecular Properties
Perhaps most exciting is the ability of AI models to infer mutation status (e.g., EGFR in lung cancer) directly from H&E images. This enables rapid PCR confirmation instead of full sequencing — cutting both time and cost significantly.
Virtual Staining
AI can generate synthetic IHC or special stains, predicting molecular and chemical properties of tissues digitally. This opens possibilities for retrospective analysis and reduces the need for additional staining procedures.
💬 2. Natural Language Processing (Text-Based AI)
While image analysis gets most of the attention, NLP is quietly revolutionizing how pathologists document, search, and learn from their work.
Speech-to-Text
AI can transcribe dictated pathology notes, learning each pathologist's vocabulary and voice patterns. This reduces transcription costs and enables real-time documentation.
Structured Report Generation
Large Language Models (LLMs) can auto-populate pathology report templates as the pathologist describes the case. This ensures consistency, reduces errors, and speeds up report generation.
Smart Search (RAG)
Retrieval-Augmented Generation enables question-answering on internal datasets (e.g., hospital or research databases) with citations to avoid "hallucinations." Pathologists can ask natural language questions like "Show me similar cases with this morphology" and get accurate, referenced results.
🔗 3. Multimodal AI (Image + Text)
The most promising frontier combines both image analysis and NLP capabilities in a single, integrated system.
Major pharmaceutical companies are already merging image analysis and NLP capabilities to accelerate drug development. This combined approach is expected to drive the next wave of breakthroughs in computational pathology because it links visual and textual insights in a single workflow.
Imagine a system where a pathologist can:
- View a slide with AI-highlighted regions
- Dictate observations that are automatically structured into a report
- Query similar cases from both image and text databases
- Get AI-suggested diagnoses based on both visual patterns and clinical history
Key Takeaway
AI + Pathologist > AI alone or human alone. The future of pathology isn't about replacing pathologists — it's about creating powerful tools that augment their expertise, making workflows faster, cheaper, and more accurate.
Looking Forward
The most promising frontier is multimodal, retrieval-augmented systems that combine images, reports, and molecular data. These systems will enable pathologists to work more efficiently while maintaining the highest standards of accuracy.
At NanoView, we're building the foundational platform that makes these AI capabilities possible. By creating infrastructure that seamlessly handles both high-resolution images and structured text data, we're positioning ourselves to enable the next generation of computational pathology tools.