Clinical Validation & Co-Development of Synthetic Cohorts in Oncology

1. Project / Cooperation Goal

We are seeking clinical research partners, academic hospitals, and imaging CROs to conduct joint retrospective validation pilots on real-world oncology datasets (specifically NSCLC clinical records and baseline CT scans).Our primary goal is to evaluate and refine CohortForge AI’s generative and reconstruction metrics in a live, clinician-governed environment.

2. What We Bring to the Project (Our Contribution)

• Deep-Tech Generative AI Studio (TRL 4 MVP): CVAE for clinical table synthesis and 2.5D residual U-Net for CT slice reconstruction.

• Fail-Closed Governance & Security: Built-in 29-field data crosswalk, subject-level split controls, and aggregate-only exports ensuring data privacy (GDPR compliance).

• Quantitative Evaluation Harness: Automated scoring for statistical fidelity (KS distance), subgroup balancing, protocol harmonization (kVp), and radiomic biomarker stability (GLCM Joint Entropy).

• HPC & Cloud Deployment Expertise: Proven experience running GPU workloads on CINECA Leonardo supercomputing infrastructure and NVIDIA Inception ecosystem support.

3. What We Are Seeking from Partners

• Clinical & Hospital Partners:

  • Data Access: Approved, secure access to retrospective thoracic oncology clinical records and baseline CT scans.

  • Clinical Guidance: Expert input from clinicians and radiologists to define key clinical variables, missingness thresholds, and plausibility criteria.

  • Validation: Independent testing of synthetic data fidelity, missing-field completion, and subgroup balance on held-out patient cohorts.

• EU Consortium Proposals (Horizon Europe / EIT Health / Digital Europe):

  • Seeking to join upcoming funding calls as the technology development & data-qualification partner to provide synthetic cohort generation and dataset harmonization.

• Imaging CROs & MedTech Industry:

  • Design-partner collaborations to test protocol-aware harmonization and cohort-qualification workflows for commercial AI pipelines.