Industrial Domains
Healthcare
The Healthcare industrial domain seeks to address the need for integrating AI technologies to improve efficiency, accuracy, and patient outcomes, while tackling challenges such as rising costs, workforce shortages, and increasing patient volumes. The main challenge is integrating AI into existing clinical workflows and electronic health record systems, addressing interoperability, data privacy, and ethical concerns. Careful validation of AI algorithms and strict adherence to regulatory frameworks are essential to ensure equitable healthcare delivery. Overcoming these challenges will allow AI to enhance clinical decision-making and transform healthcare in a responsible and sustainable manner.
Project: Context-Aware Cardiac Monitoring with On-Device AI and Federated Learning
Third parties involved: University of Ljubljana and Zentrix
This project proposes a context-aware wearable system for real-time cardiac event detection, which combines ECG, PPG, and IMU data to train a multimodal AI model via federated learning. The chest-mounted prototype features ECG and PPG sensors, which, combined with mobile IMU data, enable robust and reliable detection of cardiac anomalies. By correlating physiological signals with user activity, we reduce false alarms and improve detection accuracy. On-device federated training ensures personalised, privacy-preserving, and power-efficient real-time data analysis. The project outcome will be a TRL 4–5 validated prototype and two peer-reviewed publications, enhancing EU capabilities in preventative digital healthcare and secure, user-centric AI.
Project: DeepSense: Context-Aware AI solution for Monitoring of Patient Risks
Third parties involved: MANITTY
DeepSense aims to advance remote cardiac monitoring solution by developing a portable, non-invasive and comfortable device with embedded AI for real-time, context-aware detection of conditions such as atrial fibrillation, hypertension, and glycemic dysregulation. Fusing multimodal data (ECG, EMG, PPG, temperature, accelerometry) with on-device learning, it will enable continuous, and ultra-personalized surveillance and early intervention. Over 6 months, the project will deliver a TRL 6 working prototype (hardware, platform & AI models ) aligned with ENFIELD’s mission to foster a European AI ecosystem and support the digital transformation of healthcare through smart, interoperable medical technologies.
Selected project: Gecko: VR Healthtech App for AI driven Cardio Fitness Coach
Third parties involved: Grassroot Arts and Research UG.
The use of AI and wearables to unobtrusively monitor Parkinson’s Disease (PD) symptoms, enabling clinicians to offer personalized treatment recommendations. The focus is on detecting symptoms like tremors, bradykinesia, and dyskinesia. This project aims to develop a VR health technology prototype featuring an AI-driven Dance Fitness Coach designed to support patients with chronic cardiovascular disease by promoting physical activity and pain relief through interactive social dancing anytime, anywhere. Scientific evidence shows that dancing improves pain management, emotional well-being, mobility, and fosters social interaction. Codenamed Gecko—will collaborate with cardio physiotherapists from Riehler Gesundheitszentrum in Cologne, Germany, to train the AI agent. No clinical trials with patients will be conducted at this stage. Instead, the focus will be on developing a next-generation AI cardio fitness coach prototype integrated into GAR’s existing “Magic Plant World” VR environment, previously validated in lab tests. Through Gecko, GAR aims to redefine digital health, offering an innovative AI/VR approach to cardiovascular fitness and pain relief that goes far beyond conventional cardio exercise apps.
Selected project: Generative LLM for Single-Cell Transcriptomics of Alzheimer’s Disease
Third parties involved: LIBRA AI Technologies
Alzheimer’s disease (AD) is characterized by amyloid plaques and tau tangles, leading to brain degeneration. Traditional bulk sequencing fails to capture the cellular diversity of the CNS, but single-cell sequencing provides granular insights into individual cells. The proposed scGPT model, pre-trained on over 10 million cells, utilizes Transformers to enhance tasks such as batch correction, cell type annotation and gene network inference. The specialized scGPT-AD model aims to integrate scRNA-seq with other omics data to identify cellular perturbations and biomarkers. This approach supports personalized medicine by tailoring treatments based on individual genetic profiles, advancing our understanding and management of AD.
Selected project: Explainable association of drug side effects to changes in the proteome
Third parties involved: Cytocast
This project develops explainable machine-learning models that connect drug-induced changes in protein complexes to human side effects, using Cytocast’s Digital Twin Cell™ simulations across multiple human tissue types. By simulating hundreds of drugs and training models on the resulting interaction patterns, we aim to deliver biologically grounded explanations of safety risks. The outcome will enable European pharmaceutical companies—large and small—to perform in-silico safety testing of compounds, accelerating drug development while improving interpretability and confidence in preclinical decisions.
Project: Risk Assessment and Decision Support for ICU Readmission Prediction
Third parties involved: Paola Berchialla; Alessia Visconti; Emanuel Koumantakis – University of Torino
This project develops an ensemble of deep learning (DL) predictive models to assess ICU readmission risk using MIMIC-IV and eICU data. We will review multimodal DL models relevant to decision support, train models on MIMIC-IV, fine-tune them on eICU via transfer learning, and combine their predictions with Bayesian Model Averaging to achieve probabilistic forecasting for clinical decision-making.