Vision and Scope
Perioperative complications remain a relevant challenge. Traditional risk scores capture static information and often overlook rapid physiological changes. AISPM pursues a holistic, data-driven approach in which multimodal data and adaptive reasoning models support anaesthesiologists across the perioperative pathway. The goal is safer, more efficient, and more transparent care, always with the clinician in control.
Core Research Areas
AISPM Virtual Assistant (AISPM-VA)
A multi-agent AI platform purpose-built for perioperative medicine. It combines:
- Large Language Models (LLMs) for clinical text interpretation and dialogue,
- Predictive models (e.g., Temporal Fusion Transformers) for vital-sign forecasting and risk signals,
- Graph-based methods for drug–drug interaction context, and
- Retrieval-Augmented Generation for guideline- and evidence-aware recommendations.
The integration provides continuous, context-aware support and concise, interpretable outputs at the point of care.
Explainability and Clinical Trust
We study methods for transparent reasoning (e.g., rationale display, data provenance, uncertainty) and user-centred interaction, so that recommendations are understandable and align with clinical decision-making.
Data Science and Validation
Using more than 150,000 multimodal perioperative cases (2014–2024), we evaluate models with robust metrics (e.g., AUROC, PR-AUC, calibration, MAE). Retrospective analyses are complemented by expert review and pilot integration activities to assess usability and workflow fit.
AI Infrastructure, Ethics, and Regulation
Development runs on Inselspital’s on-premise GPU infrastructure, utilizing local data processing, in accordance with HRA and GDPR. We align with ISO 14155, SPIRIT-AI, and CONSORT-AI for clinical investigations and reporting. Data protection, auditability, and role-based access are integral to the platform design.
Collaborations and Strategic Alignment
AISPM is anchored at the Department of Anaesthesiology and Pain Medicine, Inselspital Bern, with close collaboration between clinical teams and the Graduate School of Precision Engineering (GSPRE), University of Bern (methodological and doctoral support).
The group also collaborates with the CSEM Edge & Cloud Software Group in a consulting capacity on secure and scalable AI infrastructure.
In addition, our activities are aligned with the objectives of the submitted Horizon Europe proposal CARE2CONNECT (end-user–driven generative AI in healthcare), and we have prepared materials in the Innosuisse Innolink format for potential future innovation work. (Outcomes pending; no funding decisions implied.)
Impact and Outlook
AISPM investigates how transparent AI can strengthen clinical judgement, surface relevant signals earlier, and improve situational awareness in the operating room and perioperative units.
Next steps include broader pilot evaluations, scalable deployment within secure hospital IT environments, and ongoing education for clinical users to ensure the safe and effective adoption of this technology.
