Acutelines Data Science

Group

Acute diseases such as sepsis remain a major challenge in emergency and critical care because of their rapid onset, complex mechanisms, and high variability between patients. To address this complexity, we combine advanced data science and clinical insight to develop explainable multimodal AI models that integrate heterogeneous molecular, physiological, and clinical data.

Our research leverages large-scale datasets from the Acutelines research biobank, including multi-omics profiles, waveform signals (ECG/PPG), clinical records (vital signs, lab parameters, EHRs), medical imaging, and population cohorts. We aim to identify actionable biomarkers and mechanistic pathways, and develop predictive AI systems that enable early recognition, risk stratification, personalized interventions, and real-time decision-making in clinical care. Our current research projects include:

Early recognition of high-risk sepsis patients in the emergency department using multi-omics data to identify biomarkers and mechanistic pathways.
Deep learning analysis of physiological waveforms (ECG and PPG) to predict patient severity and clinical deterioration.
Facial image analysis for non-invasive prediction of disease severity and progression.
Multimodal AI for distinguishing bacterial and viral infections using laboratory, vital sign, and waveform data.
Development of a comprehensive, explainable AI system that integrates multimodal information for early detection of clinical deterioration and supports individualized treatment strategies in sepsis care.

Through these projects, we aim to transform multimodal data into clinically actionable insights and develop explainable AI systems that can be trusted and implemented in real-world healthcare.

Relevance

How our research benefits society

Acute diseases such as sepsis require rapid and accurate clinical decisions, yet current tools often fail to capture the complexity of patient physiology and disease progression. By integrating molecular, physiological, and clinical data, our research provides a comprehensive and real-time view of patient status, improving clinicians’ ability to anticipate and respond to critical changes.

The explainable multimodal AI systems developed by our group aim to support early recognition, optimize triage, and guide personalized interventions in emergency and critical care. These models help clinicians interpret complex data, predict deterioration before it occurs, and tailor treatments to individual patients.

Ultimately, our work contributes to a more precise, data-driven healthcare system, improving patient outcomes while reducing diagnostic uncertainty and the burden on healthcare providers.