AI pathology interpretation (computational pathology)

01Overview and value chain

Markers: [EC: FDA De Novo / Breakthrough Device pathway + EU CE-MDR + IMI BIGPICTURE repository | OECD: Biotech & health | Regulator: FDA (USA), EMA (EU), NMPA (China)]

Computational pathology applies deep learning to gigapixel whole-slide images (WSIs) of stained tissue, up to 100,000 by 100,000 pixels, to automate the microscopic reading that pathologists perform manually. As the global pathologist workforce shrinks against rising biopsy volumes, AI decision- support systems take over routine tasks (mitosis counting, lymph-node metastasis search, Gleason grading of prostate cancer) and compress time to diagnosis from days to minutes, while standardizing reports and flagging suspicious foci as heatmaps. A second frontier predicts hidden molecular genetics (microsatellite instability MSI, BRAF, EGFR mutation status, PD-L1 expression) directly from cell morphology, sparing a sequencing round. The US leads commercialization under the FDA De Novo and Breakthrough Device pathways, the EU builds the 3-million-slide IMI BIGPICTURE training repository under CE-MDR, and China scales cloud telepathology for cervical, oesophageal and lung-cancer screening. The six organizations span AI software (Paige.AI, PathAI, Owkin), scanner and cloud workflow (Philips), and population-scale screening AI (Tencent Medical AI Lab, Infervision).

Key directions of AI pathology interpretation:

1. Cancer detection and grading: CNN and vision-transformer models localise micro-metastases and grade tumours (Gleason score) with pathologist-level concordance.
2. Predictive molecular pathology: morphology-only inference of MSI, BRAF, EGFR and PD-L1 status, avoiding a separate sequencing or IHC round.
3. Attention multiple-instance learning: whole-slide classification trained from only a slide-level diagnosis label, with no per-cell annotation.
4. Digital pathology workflow and telepathology: high-throughput scanners and cloud LIS integration stream cases to remote pathologists.

Sectoral value chain

[glass slide] ──> [WSI scanner] ──> [tiling & feature extraction (ViT)]
                                                          │
                                          (attention-MIL aggregation)
                                                          │
                                                          ▼
[pathologist sign-out in LIS] <─── [tumour heatmap & report]

Value chain levels

Cross-cutting technologies of the sector:

• whole-slide-imaging: gigapixel digital slides (up to 100,000x100,000 px at x40) scanned at 0.25 micron/pixel for nuclear-chromatin review.
• attention-multiple-instance-learning: weakly supervised slide classification that learns from a single slide-level label by attention-weighting thousands of patch embeddings.
• self-supervised-vit-encoders: contrastive self-supervised vision transformers that produce reusable 1024-dim patch embeddings without manual annotation.

02US

The United States leads FDA-cleared AI pathology, with the first De Novo clearances and the PathAI biomarker-prediction platforms.

Paige.AI FDA-first, PathAI immunotherapy models, CAP/CLIA standards

• Paige.AI (New York, spun from Memorial Sloan Kettering): built the first AI pathology product cleared by the FDA (Paige Prostate), and in 2025 received FDA Breakthrough Device designation for Paige PanCancer Detect, an AI tool that flags suspicious cancer foci across multiple tissues and organs on a single AI-first digital-pathology platform.
• PathAI (Boston): its Path-IO (Pathology-driven Immunotherapy Optimization) multimodal platform combines pathomics, radiomics and clinical data to predict immunotherapy response in non-small-cell lung cancer, and its PathAI Discovery platform supports PD-L1 and stromal TIL biomarker development for clinical trials.
• CAP/CLIA validation: the College of American Pathologists drives algorithm-validation standards for clinical CAP/CLIA laboratories, and US venture capital funds the consolidation of AI-pathology startups.

03CN

China pursues population-scale digital-pathology deployment over the cloud, anchored by the Healthy China programme and hyperscale AI screening models.

Tencent automated cytology, Infervision lung-nodule AI, Healthy China

• Tencent Medical AI Lab: published a fully automated AI slide-check system in Nature (January 2026) that performs multi-focal whole-slide scanning for 3D cell information and auto-identifies abnormal cells, validated on 1,124 cervical liquid-based cytology samples across four centres with high concordance to human readers; its Miying imaging AI also screens oesophageal and lung cancer.
• Infervision (Beijing, founded 2016): its InferRead AI performs sub-millimetre lung-nodule detection (below 5 mm nodules, ground-glass and vessel-convergence features) with individualized malignancy-probability scoring, deployed across Chinese hospitals along the patient pathway from screening to follow-up.
• Healthy China: the national programme connects regional hospitals to centralized AI pathology servers in Beijing and Shanghai, and Chinese vendors lead in ultra-fast robotic slide scanners.

04EU

The European Union couples strict GDPR and CE-MDR oversight with the world’s largest annotated WSI repository and leading scanner-and-cloud workflow vendors.

Philips IntelliSite cloud, Owkin federated learning, BIGPICTURE

• Royal Philips (Netherlands, NYSE: PHG): in March 2026 expanded its digital-pathology portfolio with cloud-enabled Philips IntelliSite Pathology Solution on HealthSuite (powered by AWS), integrating high-throughput scanners with LIS connectivity for hospital pathology departments across Western Europe.
• Owkin (France): a federated-learning AI company that, in a Nature Medicine report, trained deep-learning models on multi-hospital histopathology data without sharing patient slides, and pursues generalizable pathology, spatial transcriptomics and agentic AI co-pilots for researchers.
• BIGPICTURE (IMI): the Innovative Medicines Initiative consortium assembles the largest annotated digital-slide repository (more than 3 million WSIs) for training European AI models under GDPR and CE-MDR.

05Leading companies and research institutes

06Tech stack and innovations

The stack pairs high-resolution digital pathology with self-supervised deep learning and weakly supervised whole-slide classification.

1. Whole-slide imaging and preprocessing:
   • an H&E-stained biopsy slide is scanned on a Leica Aperio GT 450 or Philips IntelliSite at x40 (0.25 micron/pixel) into a 1-3 GB .svs file; OTSU thresholding segments tissue from background, and the tissue is tiled into 512x512 patches for GPU processing.
2. Self-supervised vision-transformer encoding:
   • a contrastive self-supervised vision transformer (ViT) pretrained on millions of unlabelled patches maps each 512x512 tile to a 1024-dimensional embedding, capturing nuclear-chromatin texture without manual annotation.
3. Attention-MIL aggregation and heatmap:
   • an attention pooling head weights each patch embedding to form a slide-level vector, from which classifiers output tumour probability and molecular-biomarker predictions, and a pixel-level heatmap is overlaid on the slide for the pathologist to review in QuPath or Aperio ImageScope.

07Value chains and production pipelines

Industrial pipeline of AI WSI interpretation (CAP/CLIA + CE-MDR compliant)

┌───────────────────────────┐      ┌───────────────────────────┐
│ 1. WSI scanning           │ ───> │ 2. Tissue segmentation    │
│    (Aperio GT 450, x40)   │      │    & 512x512 tiling       │
└───────────────────────────┘      └───────────────────────────┘
                                                  │
                                                  ▼
┌───────────────────────────┐      ┌───────────────────────────┐
│ 4. Attention-MIL          │ <─── │ 3. ViT feature extraction │
│    aggregation            │      │    (1024-dim embeddings)  │
└───────────────────────────┘      └───────────────────────────┘
              │
              ▼
┌───────────────────────────┐      ┌───────────────────────────┐
│ 5. Tumour heatmap &       │ ───> │ 6. Pathologist sign-out   │
│    biomarker prediction   │      │    in LIS                 │
└───────────────────────────┘      └───────────────────────────┘

Stage 1: Whole-slide imaging

The H&E-stained section is scanned on a Leica Aperio GT 450 or Philips IntelliSite scanner at x40 objective (0.25 micron/pixel), producing a 1-3 GB gigapixel .svs file that preserves nuclear-chromatin detail.

Stage 2: Tissue segmentation and tiling

An OTSU threshold strips the white background and glass edges, isolating tissue, which is then divided into a grid of 512x512-pixel patches sized for GPU memory.

Stage 3: Self-supervised feature extraction

Each patch is passed through a contrastive self-supervised vision-transformer encoder, yielding a 1024-dimensional embedding vector that encodes morphology; the encoder is reused across cancer types without retraining.

Stage 4: Attention-MIL aggregation

An attention pooling head computes an importance weight for every patch embedding and combines them into a single slide-level feature vector, enabling weakly supervised classification from a slide-level diagnosis label alone.

Stage 5: Tumour heatmap and biomarker prediction

Classifiers on the slide vector output tumour probability, Gleason grade and predicted MSI/BRAF/EGFR/PD-L1 status, and a pixel-level heatmap of suspicious foci is generated for overlay on the original slide.

Stage 6: Pathologist sign-out in LIS

The pathologist opens the WSI with the AI heatmap in a viewer (QuPath, Aperio ImageScope), confirms or corrects the findings, and signs out the final report in the laboratory information system with the AI output co-displayed.