Task
In a medical measurement system, electrodes are used to measure an electrochemical reaction. During clinical tests, these sensors exhibited problems that could lead to incorrect measurements and which must therefore be detected by the measurement system's software in future. As part of this project, methods were developed to detect signs of these errors in the sensor's signal curve in good time. This allows faulty sensors to be taken out of service before incorrect measurements can occur.
Solution approach
- Complex time series and correlation analyses
- Development of criteria for detecting faulty curve progressions
- Pre-development of algorithms in Matlab
- Parameter optimisation for the models
- Implementation of signal processing models in Simulink
Result
The figure shows the ROC curve for one of the developed error detection mechanisms. Two models were proposed for this purpose. For each set of model parameters, a point (sensitivity–specificity pair) can be calculated on the graph.
Projektteam
Dr. Irina Ostapenko
Algorithms, data analysis and organisation
Dr. Momme Winkelnkemper
Algorithms, data analysis and modelling
