The development of rapid and accurate biomedical laser spectroscopy systems in the mid-infrared has been enabled by the commercial availability of external-cavity quantum cascade lasers (EC-QCLs). EC-QCLs are a preferable alternative to benchtop instruments such as Fourier transform infrared spectrometers for sensor development as they are small and have high spectral power density. They also allow for the investigation of multiple analytes due to their broad tuneability and through the use of multivariate analysis. This article presents an in vitro investigation with two fiber-coupled measurement setups based on attenuated total reflection spectroscopy and direct transmission spectroscopy for sensing. A pulsed EC-QCL (1200–900 cm−1) was used for measurements of glucose and albumin in aqueous solutions, with lactate and urea as interferents. This analyte composition was chosen as an example of a complex aqueous solution with relevance for biomedical sensors. Glucose concentrations were determined in both setup types with root-mean-square error of cross-validation (RMSECV) of less than 20 mg/dL using partial least-squares (PLS) regression. These results demonstrate accurate analyte measurements, and are promising for further development of fiber-coupled, miniaturised in vivo sensors based on mid-infrared spectroscopy.

A fiber-coupled transmission spectroscopy setup using a pulsed external-cavity quantum cascade laser (EC-QCL, 1200-900 cm −1 ) has been developed and demonstrated for measurements of aqueous solutions. The system has been characterised with regard to the laser noise and optimal optical pathlength. Solutions with glucose were used to further test the setup, and glucose concentrations down to physiologically relevant levels (0-600 mg/dL) were investigated. Albumin, lactate, urea, and fructose in various concentrations were added as interfering substances as their absorption bands overlap with those of glucose, and because they may be of interest in a clinical setting. Analyte concentrations were predicted using partial least-squares (PLS) regression, and the root-mean-square error of cross-validation for glucose was 10.7 mg/dL. The advantages of using a convolutional neural network (CNN) for regression analysis in comparison to the PLS regression were also shown. The application of a CNN gave an improved prediction error (8.3 mg/dL), and was used to identify important spectral regions. These results are comparable to state-of-the-art enzymatic glucose sensors, and are encouraging for further research on optics-based glucose sensors.