Photonic measurement techniques, including vibrational spectroscopy, are increasingly being used in various disciplines such as life sciences and medicine. This increased use is linked to improvements in measurement methods and instrumentation, but also to the development of data science methods and computational infrastructures. These photonic data science methods enable the detection and extraction of high-level information from subtle differences in (biomedical) vibrational spectra and images. The high-level information depends on the task and the sample. For example, the prediction of tissue types, disease states or sample properties such as concentrations of constituents could represent such high-level information.
Vibrational spectroscopic techniques such as Raman spectroscopy and IR spectroscopy and related imaging techniques have several advantages, for example they can be used as non-destructive fingerprinting techniques. In order to exploit the full potential of these vibrational fingerprints, the entire life cycle of the spectroscopic data, from its generation to data modelling and archiving, is important and needs to be studied in a holistic way. In particular, experimental design, sample size planning, data pre-treatment, data pre-processing, chemometric and machine learning based data modelling, model transfer methods and transfer learning are important. All procedures are sequentially combined in a data pipeline that standardises the vibrational data and extracts reliable high-level information.
Here, we present our recent studies aimed at using AI-based techniques such as machine learning and deep learning models to solve the inverse problem of optical measurement techniques aimed at standardising the optical data. This standardised data can be used to extract high level information from the samples. Here we will discuss studies on how Raman spectra and (spectroscopic) image data can be used to predict higher-level information such as analyte concentrations or diagnostic markers.
Acknowledgement:
Financial support of the EU, the ‘Thüringer Ministerium für Wirtschaft, Wissenschaft und Digitale Gesellschaft’ (TMWWDG), the ‘Thüringer Aufbaubank’, the Federal Ministry of Education and Research, Germany (BMBF), the German Science Foundation, the Fonds der Chemischen Industrie, the Carl-Zeiss Foundation and Leibniz association is greatly acknowledged.
