Csfometer: A continuous cerebrospinal monitorization system

Project no. 29676 · Status: Kapanmış · 2023

Summary

Objective: This study aimed to develop a proof-of-concept, low-cost, non-invasive device (BOSoMetre) capable of continuously monitoring cerebrospinal fluid (CSF) color and turbidity in external ventricular drainage (EVD) systems. The purpose was to predict infection onset and clearance, potentially optimizing the timing of CSF sampling and reducing unnecessary specimen collection. Methods: The device was developed using an Arduino microcontroller and TCS3200 color sensor, measuring CSF color changes through spectrophotometric techniques. The system includes LED illumination optimized with a 68kΩ resistor, real-time data logging, SD card storage, and uninterrupted power supply. The optical components were housed in a custom 3D-printed dark chamber and integrated into the EVD line. Sensor outputs were optimized to record data every minute, with red, green, blue, and clear channel measurements taken for each recording. Clinical evaluation prospectively monitored patients with EVD, with three CSF samples collected weekly per standard protocol. CSF infection clearance was defined according to Infectious Diseases Society of America (IDSA) 2017 criteria. Two machine learning approaches were tested using the Subspace KNN algorithm: a binary classification (infected/non-infected) and a triphasic model that included an intermediate "gray zone" between confirmed states. Results: Analysis of 1,800,000 recordings from 12 patients demonstrated the device's capability to detect CSF infections with high accuracy. The binary classification model, using an 80-20 train-test split, achieved 98.9% sensitivity and 97.1% specificity in detecting infected CSF samples. When implementing a confidence threshold for the binary model, samples below the threshold were classified as indeterminate, resulting in 22% sensitivity for these ambiguous cases while maintaining high accuracy for definitive cases. The triphasic model, while better at identifying ambiguous cases (64.3% sensitivity for intermediate zone), showed decreased sensitivity for infected cases, making it less suitable for clinical implementation where missing infections could have serious consequences. Conclusion: BOSoMetre demonstrates potential as a low-cost, continuous monitoring system for early detection of CSF infections and optimization of sampling timing. While the binary classification model shows promising accuracy, further validation with larger patient cohorts is needed to refine the algorithm and enhance detection accuracy. The device could serve as an early warning system and help reduce unnecessary CSF sampling, particularly beneficial in resource-limited settings. Future studies should focus on prospective validation and integration into clinical decision-making protocols.

View details