ISIS program: a new tool for medical research at the bedside in critical care units

Introduction

The goal of this program is to develop an experimental tool able to record, store and analyse data issued from critical care patients. Due to technical limitations and medical constraints, information systems able to manage such data flow are difficult to deploy.

Methods

Data recording is done through a laptop connected to the medical devices, allowing analogical and digital signal transmission through a high-speed network. Several servers are dedicated to specialised tasks: mass storage, model generation, artificial intelligence (AI), telecommunications, and security. A 3 Teraflops supercomputer is dedicated to intensive computation when necessary. Twenty applications are dedicated to elective tasks, most of them running using the Linux operating system.

The 'Aiddiag' data-acquisition software is a standalone application adapted to patient data recording from the biomedical devices and caregiver's inputs. It has a friendly designed user-interface touchscreen at the bedside and was adapted according to caregivers' feedback. Data are also stored in a repository and a selective secondary extraction is possible. Online and offline analysis by the AI engine is allowed. Software had to consider time specifications and uses distributed computation to achieve high workload tasks. We complied to the French legal patient data management constraints.

Results

After 2 years, our system is fully deployed. It recorded more than 2,500 patient-hours over a 3-month period. Signal loss is less than 1%. Our tool allows recording of more than 40 digital signals, eight analogical signals sampled at a rate of 1 kHz, and caregiver comments and actions. CPU resources of the laptop are available for supplemental AI developments during data acquisition. Transfer of data to the repository is either a hotplug-automated process or delayed with 5 days of buffering in the laptop. Automated artefacts' cleaning allows time-series analysis (GARCH method) to extract behavioural models after intensive computation. The AI engine is used for medical guideline implementation (that is, severe brain trauma care algorithms) and later comparison with caregiver's behaviour. Remote use of our system is possible and schedulable, allowing other research teams to work on the data. Limitations have been detected during intensive calculation. Fine-tuning of the network will suppress these limitations.

Conclusion

ISIS is the first program to achieve an easy-to-use recording tool able to build a very large medical repository. Data analysis methods and AI-controlled automated complex medical guidelines are under evaluation.