💰 Funding Details

EIC Pathfinder Challenge: Generative-AI Agents for Cancer Care

What is funded?

The call HORIZON-EIC-2025-PATHFINDERCHALLENGES-01-02 provides lump-sum grants (up to €120 million per project) for visionary, high-risk research that will create interactive Generative-AI (GenAI) agents – or super-agents – able to deliver a holistic, end-to-end clinical pathway for one of eight priority cancers (breast, cervical, ovarian, prostate, lung, brain, stomach, colorectal).

Projects must:

* Address both Area 1 *Technological* and Area 2 *Clinical* objectives.

* Be early-stage, “beyond-state-of-the-art” concepts with clear proof-of-concept plans.

* Comply with EU ethical, privacy and Trustworthy-AI requirements and contribute assets to the Cancer Image Europe platform & EHDS catalogue.

* Fit within €0.5 – 4 million lump-sum budgets, using the Model Grant Agreement (HORIZON-AG-LS).

Who can apply?

* Single legal entities (except mid-caps & large companies) or consortia.

* Consortia of ≥2 must involve independent entities from at least two different Member States/Associated Countries; ≥3 entities must include one from a Member State and two from different your country or Associated Countries.

* Typical stakeholders: universities, RTOs, hospitals, SMEs, start-ups, patient groups, AI labs.

Key dates

* Call opens: 24 Jul 2025

* Deadline: 29 Oct 2025, 17:00 CET (single stage)

* Page limit: 30 A4 pages (Part B)

Expected impacts

1. Next-generation GenAI agents for accurate diagnosis & personalised therapy.

2. Shared synthetic imaging database & benchmarking frameworks.

3. Cost reduction and efficiency gains for EU healthcare systems.

4. Reinforced EU leadership in trustworthy medical AI.

---

📊 Evaluation Criteria

Evaluation Highlights

Excellence (40 %)

* Ground-breaking nature & ambition of GenAI concepts.

* Fit to both Areas and selected cancer.

* Quality of multidisciplinary methodology & risk management.

Impact (30 %)

* Contribution to Trustworthy AI, Cancer Plan, AI Strategy.

* Adoption potential in your country and EU healthcare systems.

* Portfolio synergies, data/tool contribution to Cancer Image Europe.

Quality & Efficiency of Implementation (30 %)

* Complementarity of partners & access to infrastructures.

* Clear workplan with milestones, metrics & uncertainty quantification.

* Sound lump-sum justification.

📝 Application Strategy

Strategic Application Guide

1. Positioning the Idea

1. Choose one cancer only. Map unmet clinical needs and how a GenAI *super-agent* will outperform current practice.

2. Articulate dual alignment:

* Area 1 – e.g. multimodal fusion + synthetic data generation + knowledge graph.

* Area 2 – e.g. predictive diagnosis agent and treatment-selection agent.

2. Build a Category-balanced Consortium

| Category | Requirement | Suggested Partner Type |

|----------|-------------|------------------------|

| Cancer | Expertise & clinical data for chosen cancer | Comprehensive cancer centre, hospital |

| Technology | Complementary AI approaches (e.g. geometric DL, graph NN, neural fields) | AI SME, university lab |

| Infrastructure | Access to imaging biobanks, HPC, regulatory know-how | RTO, standards body |

Early Letters of Commitment from clinical sites guarantee data & PoC environments.

3. Design the Scientific Workplan (WPs)

1. WP1: Data governance, ethics, Trustworthy-AI compliance, EHDS metadata.

2. WP2: Multimodal & multidimensional data integration (Area 1-a).

3. WP3: GenAI-driven synthetic data augmentation (Area 1-b).

4. WP4: Explainable super-agent for predictive diagnosis (Area 2-a).

5. WP5: Personalised treatment recommendation module (Area 2-b).

6. WP6: Clinical proof-of-concept vs. SoC (prospective/retrospective).

7. WP7: Portfolio collaboration, benchmarking, dissemination & IP.

4. Impact & Portfolio Synergy

* Quantify KPIs: AUROC ≥ 0.90, treatment-selection accuracy ≥ 20 % over baseline, 30 % clinician time saved.

* Sharing plan: synthetic datasets & models to Cancer Image Europe under agreed licences.

* Regulatory roadmap: ISO 13485, MDR classification, AI-Act conformity assessment.

5. Implementation & Budgeting Tips

* Use work-package lump-sum breakdown matching the cost-drivers (data acquisition, compute, clinical study).

* Allocate ≥5 % for ethics, legal, gender & diversity monitoring.

* Reserve travel & PMs for portfolio-wide workshops mandated by EIC.

---

💡 Success Factors

Early Clinical Validation Path

Design a **proof-of-concept study** (prospective or retrospective) embedded in WP6, with ethics approval timeline, patient numbers, endpoints and success metrics clearly stated.

Robust Data Governance & EHDS Compliance

Implement **FAIR principles**, Health-DCAT-AP metadata, GDPR-compliant de-identification and secure federated learning where needed.

Comprehensive Trustworthy-AI Framework

Embed bias assessment, explainability dashboards, human-in-the-loop controls and alignment with the upcoming **AI Act risk-management system**.

Scalable Synthetic Data Pipeline

Show how GenAI will generate **high-fidelity, labelled synthetic images and -omics data** that augment under-represented cohorts and are shareable via Cancer Image Europe.

Interdisciplinary Consortium Synergy

Include oncologists, radiologists, AI scientists, ethicists, patient advocates and regulatory experts with **clearly delineated roles and decision processes**.

Clear Exploitation & Sustainability Plan

Outline **post-grant commercialisation**, CE-marking steps, and long-term hosting of agents on clinically integrated platforms.

❌ Common Mistakes to Avoid

Ignoring the Dual-Area Requirement

Projects that focus **only** on technological enablers *or* clinical use cases are declared out of scope. Every proposal must tackle **at least one sub-objective in Area 1 _and_ Area 2**.

Selecting Non-eligible Cancer Types

Including pancreatic or liver cancer, for example, immediately renders the proposal ineligible. Stick strictly to the **eight listed cancers**.

Black-Box AI Without Interpretability Plan

Evaluators penalise models that do not incorporate **explainability, causal inference or uncertainty quantification** consistent with Trustworthy-AI guidelines.

Insufficient Diversity & Gender Sensitivity in Data

Failing to demonstrate how datasets will cover **age, gender, ethnicity** variations—and how synthetic data will mitigate bias—lowers both Excellence and Impact scores.

Overlooking Portfolio Collaboration Duties

The call states that projects must **share synthetic data, benchmark agents and cross-validate** at each other’s clinical sites. Omitting resources for these tasks is a critical flaw.

Mismatch with Lump-Sum Logic

Submitting a traditional cost-reimbursement budget instead of a **work-package-based lump-sum** breakdown leads to administrative rejection.

Generative-AI based Agents to Revolutionize Medical Diagnosis and Treatment of Cancer

Quick Facts for Spain Applicants

Ready to Apply from Spain?