AI Model Training
3M+ pathology images. Matched molecular data. One source.
3M+
unique digital pathology images across H&E and IHC modalities
50+
states represented — community-setting data, no academic bias
100%
from the source. Every image generated from slides at NeoGenomics laboratories
NeoPixel is the data foundation for a range of AI/ML and biomedical research programs, from algorithm training through to clinical validation and regulatory submission support.
AI/MLAlgorithm training and validation
Train and validate biomarker prediction algorithms utilizing a wide oncology image library matched with molecular data and medical records for model training and evaluation.
- Diverse, community-representative patient population minimizes training bias
- Matched IHC scores and NGS results for supervised learning ground truth
- Scanner type and magnification metadata for cross-platform validation
Digital Biomarker DevelopmentFrom slide image to digital biomarker
Identify and validate novel digital biomarkers from H&E morphology — tumor microenvironment features, spatial patterns, and image-derived signatures — using archival samples with rich clinical context.
- H&E images linked to downstream treatment and outcome data
- Multi-tumor cohorts for pan-cancer feature discovery
- Archive access for prospective validation against new hypotheses
ResearchBiomarker correlation studies
Investigate associations between image-derived features and molecular data, IHC expression levels, mutation status, and gene fusions, using linked multi-modal data from the same specimen.
- IHC and NGS data linked at the patient and specimen level
- Rare tumor cohorts available for niche research programs
- Retrospective archive enabling hypothesis validation without new collection
CDx DevelopmentCompanion diagnostic algorithm support
Support companion diagnostic development and regulatory submissions with a clinical-grade image library, including diverse population representation required for FDA Algorithm authorization.
- Demographic diversity across 50 states, >85% community setting
- Structured scanner and staining metadata for cross-site generalization
- Clone-level IHC scoring for CDx-specific training sets
What comes with every image
Not training imaging models? Our unique multimodal data pipeline gives AI/ML teams the cross-modal signal needed to train models that reflect the biological complexity of cancer, with matched data that can connect morphology to molecular status, reduce academic-center bias through broad community representation, and support defensible research with first-party provenance. With 500,000+ new patient records generated annually, NeoGenomics provides the scale, diversity, and traceability required for model pre-training, validation, and future clinical deployment.