Dandelion Health

Query Dandelion for patients with AFib history.

Extract paired 12-lead ECGs and EHR diagnosis codes.

Train CNN model in the secure sandbox.

Validate against a held-out diverse geographic cohort.

Define the 'Pivotal Study' cohort parameters.

Access Dandelion's independent validation set.

Run the model via API on the validation data.

Generate automated performance reports for FDA review.

Identify patients with Stage III Lung Cancer.

Aggregate all CT, PET, and pathology scans over a 5-year window.

Analyze response to specific immunotherapy regimens.

Identify early biomarkers of drug resistance.

Subset existing model performance by Dandelion's demographic tags.

Identify performance gaps in rural health system data.

Fine-tune the model using Dandelion's targeted diverse datasets.

Certify the model as 'Bias-Verified'.

Access Dandelion's repository of de-identified pathology slides.

Link slides to final pathology reports for ground-truth labeling.

Train segmentation models on specific cell types.

Validate accuracy against expert pathologist annotations provided.

Search across the multi-million patient database for rare ICD-10 combinations.

Extract clinical notes and lab results.

Train NLP models to identify 'suspicious' patterns in undiagnosed patients.

Deploy screening tools to partner health systems.

Generate synthetic patient cohorts.

Use Dandelion's real-world data as a 'fidelity baseline'.

Run statistical parity tests between synthetic and real waveforms.

Validate that models trained on synthetic data perform identically on Dandelion's real-world data.