The exponential growth of wearable healthcare devices market is fostered by the internetof-things (IoT) era. Connected smart biosensors enable a decentralized healthcare thatdoes not constrain the user to be in a medical facility to get a real-time insight on hishealth status and a medical diagnosis from a doctor. Moreover, remote physiologicalmonitoring is appealing in sport applications where athletes need real-time feedback ontheir level of dehydration and muscle fatigue in order to optimize their performances.Electrochemical sensors play a crucial role in physiology and healthcare monitoring sincethey provide information at molecular level, where the biosensor is in direct contactwith bodily fluids such as sweat. A comprehensive healthcare diagnosis is achieved bycontinuously monitoring several types of biomarkers because of correlations betweenbiological compounds. Namely, endogenous metabolites such as lactate, or potassium andammonium ions, enable the quantification of muscle fatigue, hence, preventing musclecramping. In therapeutic drug monitoring, exogenous compounds are continuously trackedso that the drug is maintained in its therapeutic range, in order to always be effectiveand not toxic for the patient. Besides multi-sensing and general-purpose capabilities,electrochemical platforms need to be correlated to the health and physiological statusof the user, where the large amount of measured biological data must be accuratelyprocessed and interpreted by smart data analytic tools.This thesis covers the design, implementation, characterization, and validation of hardwareand software interfaces for multi-panel electrochemical sensing platforms.A multi-mode hardware front-end enabling voltammetric and potentiometric measurementsis designed to provide a continuous and concurrent monitoring of endogenousmetabolites, drugs, and electrolytes. This versatile and multi-sensing platform offers aportable solution for remote and comprehensive healthcare monitoring.Moreover, a multi-ion-sensing front-end is designed for accurate physiology in sweatsensingapplications. The hardware is proposed as a solution for multiple electrolytedetection in artificial sweat samples. In such complex media, multi-ion-sensors are subjectto interference from background electrolytes that considerably distorts sensor response.Therefore, a compact and analytical model of ion-sensing transduction mechanism isproposed to understand both qualitatively and quantitatively the non-linearity inducedby these artifacts. The ion-sensor model is implemented at the core of an emulator ofsynthetic datasets that is built to simulate ion-sensor responses in artificial sweat samples.The emulator addresses the expensive time and chemical resources needed to acquire largedatabase for training multivariate calibration models. Thus, the emulated data is used forthe training and optimization of a multi-output support vector regressor that is proposedas an accurate, unbiased, robust, compact, low-complexity, and low-latency estimatorfor the multivariate calibration of multi-ion-sensors. Then, the multi-ion-sensing array,the analog front-end interface, and the chemometric model deployed on a RaspberryPi, are seamlessly co-integrated for the monitoring of sodium, potassium, ammonium,and calcium ions in artificial sweat, within an IoT framework for real-time and accuratephysiology.
EPFL2021
