BRAINTWIN

In neurology, where disease heterogeneity limits conventional methods, digital twin technology offers a paradigm shift. For primary brain tumors (gliomas, meningiomas and primary central nervous system lymphomas) and cerebral small vessel disease (SVD), conditions characterized by unpredictable courses and variable treatment responses, BRAINTWIN proposes an innovative architecture for generating personalized digital twins.

Current approaches, based on population statistics and a limited number of biomarkers, do not capture individual disease trajectories. No tool currently exists to estimate the risk of stroke recurrence or cognitive decline in SVD, and the personalized management of gliomas and lymphomas remains challenging because of their molecular heterogeneity. To simulate disease evolution and treatment response, BRAINTWIN combines AI-driven modelling with multimodal data fusion, including neuroimaging, histopathology, clinical data and omics data.

Four key innovations structure the project infrastructure: explainable multimodal fusion based on foundation models; federated learning that is resilient to attacks and compliant with the GDPR; generative modelling using Schrödinger bridges to model longitudinal disease trajectories and complete missing modalities; and cross-domain clinical validation in neuro-oncology, SVD and neurodegeneration.

The project is organized into seven work packages covering data integration, AI model development, federated infrastructure, generative modelling, clinical validation, regulatory compliance and dissemination. Expected outcomes include new algorithms for multimodal fusion, robust federated learning and trajectory generation; clinically validated digital twins for precise stratification and earlier interventions; and a framework compliant with European regulations, strengthening trust and data protection.

By combining cutting-edge AI and clinical expertise, BRAINTWIN aims to transform personalized neurology. Its scalable framework can be extended to other diseases while addressing the challenges of privacy, interpretability and clinical applicability.