Lung cancer diagnosis and treatment require clinicians to interpret large volumes of heterogeneous patient data. At the same time, disease progression and treatment response can vary significantly between individuals.

LUMINA brings together multimodal health data and advanced AI methods to support more accurate prediction, risk assessment and personalised treatment planning.

Research Approach

The project’s interconnected work packages cover the full research pathway, from data preparation and model development to technical validation and collaboration.

LUMINA will:

• Build a diverse multimodal data foundation by collecting, curating and harmonising longitudinal lung cancer data from clinical partners in Denmark, Italy, Poland and South Korea.
• Develop the LUMINA Data Suite, a secure and interoperable framework for integrating clinical data and preparing it for AI development.
• Generate realistic synthetic data to supplement available patient data, address missing information and support privacy-conscious AI research.
• Develop the BRONCHI Digital Twin methodology using physics-informed and attention-based AI models designed to represent changes in the lungs over time.
• Model disease progression and treatment response, initially focusing on non-small cell lung cancer and immunotherapy.
• Identify patient-specific risk factors and candidate biomarkers associated with disease progression, recurrence, survival and treatment response.
• Combine physics-informed modelling with explainable AI methods to provide greater transparency about the factors contributing to individual outputs.
• Validate the BRONCHI Digital Twin prototype using multi-centre data, synthetic test scenarios and structured feedback from clinicians.
• Integrate the project’s data and AI components into an open and interoperable research platform that supports testing, benchmarking and collaboration across the EIC Pathfinder Challenge portfolio.

Beyond lung cancer

Although LUMINA initially focuses on lung cancer, its underlying methodology is designed as a modular framework.

The project’s data structures, cross-modality models, physics-informed methods and technical interfaces could potentially be adapted to other cancers with different data modalities and disease dynamics. Such transfer would require further development, training and validation for each cancer type and clinical context.

The modular design nevertheless gives LUMINA the potential to contribute to the broader development of predictive and personalised oncology.