Evaluation of a system for sepsis detection with scalable AI-platform

Important results from the project

Early detection of severe bacterial infections is of utmost importance; they have high mortality and long care periods which entail great costs for society. An IT support that continuously monitors all inpatients in a hospital would probably be a great benefit for the patient and healthcare system. The aim of the project was to make a technical evaluation of Neuron - a general AI platform - with an algoritm for detecting patients with high risk of obtaining sepsis and to investigate the CE situation, ie the path to commercialization of the technology in the healthcare sector.

Expected long term effects

A sepsis detection algorithm was developed to evaluate our high-performance computer cluster-based AI approach. The project shows that the technology is complete and robust and that system performance is fully sufficient to serve larger hospitals and regions with regards to sepsis detection. Commercialization requires a controlled clinical study with Swedish patient data and quality assurances in accordance with the EU´s new Medical Device Regulation. Our AI approach has attracted interest outside healthcare in the maritime sector and for air traffic control.

Approach and implementation

The AI algorithm for sepsis prediction - LiSep LSTM - was developed and trained on the MIMIC-III dataset in AP1 and AP2. Our high-performance computer cluster (software and hardware) was installed in RÖ´s data hall with integration with anonymised patient data sources (AP3). In AP4, the system was to be evaluated in clinical practice (clinical study) at the University Hospital in Linköping but this could not be implemented due to Covid-19. Instead, a technical test of the system´s performance was performed at RÖ. AP5 examined the path to a CE for the product.