61 60 Principal Scientist Profiles Thomas Bocklitz Thomas Bocklitz Principal Scientist Profiles HEAD OF LEIBNIZ-IPHT DEPARTMENT “PHOTONIC DATA SCIENCE” PD Dr. Thomas W. Bocklitz studied physics and earned his PhD in Chemometrics from the Friedrich Schiller University Jena, Germany. He is head of a research department “Photonic Data Science” at the Leibniz Institute of Photonic Technology. His research agenda is closely connected with the translation of physical measurements into bio-medical relevant information. Dr. Bocklitz received the Kowalski-award in 2015 and the Kaiser-Friedrich research award in 2018 together with the CDIS team. In 2024, he was appointed full professor at the Institute of Physical Chemistry at the Friedrich Schiller University Jena. THOMAS BOCKLITZ RESEARCH AREAS Dr. Bocklitz’s research interests are focused on the application of data science methods: • Machine learning (ML) for photonic image data • Chemometrics and ML for spectral data • Correlation of different measurement methods and data fusion of the measurement data TEACHING FIELDS Dr. Bocklitz teaches undergraduate courses in mathematics for chemist (B.Sc.) as well as classes in fundamental physical chemistry (M.Sc. Chemistry). He is give lectures in chemometrics for M.Sc. Medical Photonics students and workshops in the JSMC. RESEARCH METHODS The department headed by Dr. Bocklitz is theory-oriented. While the calculation equipment is constantly upgraded, computational methods are developed in-house and include: Machine learning for photonic image data: • Classical machine learning methods • Deep learning methods, like image classifiers, methods for semantic segmentation and GANs • Data preprocessing and data standardization based on inverse modelling Chemometrics and ML for spectral data: • Multivariate statistics and machine learning methods for spectral data • Deep learning models for pretreatment and spectral analysis • Data preprocessing and data standardization based on inverse modelling Correlation of different measurement methods and data fusion of the measurement data: • Data fusion of various data type combinations • Co-registration and spatial correlation of measurements RECENT RESEARCH RESULTS The Bocklitz research group investigates the entire data life cycle of photonic data, which extends from data generation to data modelling, data learning and archiving. The data life cycle is considered in a holistic approach and methods and algorithms for experiment planning, sample size planning [1], data pre-treatment and data standardization [2] are investigated. These procedures are combined with chemometric procedures [3], model transfer methods [4] and artificial intelligence based techniques [5] in a data pipeline. This holistic approach allows the use of data from various photonic processes for material characterization and medical diagnostics. Further focal points of the research department are the data fusion of different heterogeneous data sources [6], the simulation of different measurement procedures in order to optimize correction procedures, methods for the interpretation of analysis models [7] and the construction of data infrastructures for different photonic measurement data, which guarantee the FAIR principles [8]. SEMANTIC SEGMENTATION OF NON-LINEAR MULTIMODAL IMAGES FOR DISEASE GRADING OF INFLAMMATORY BOWEL DISEASE – A SEGNET-BASED APPLICATION Non-linear multi-contrast microscopy is the combination of coherent anti-stokes Raman scattering (CARS), two-photon excited fluorescence (TPEF) and second harmonic generation (SHG), and it has shown its potential to diagnose different inflammatory bowel diseases (IBDs). Non-linear multi-contrast microscopy can quantify biomolecular changes in the crypt and mucosa region, which serve as a predictive marker for IBD severity. To use the multicontrast images for IBD severity determination, an automatic segmentation of the crypt and mucosa regions must be performed. In the presented study, we semantically segment the crypts and the mucosa region using a deep neural network, i.e. a SegNet. The semantic segmentation based on the SegNet architecture is compared to a classical machine learning approach, which revealed that the trained SegNet model achieved an overall F1 score of 0.75 and outperformed the classical machine learning approach for the segmentation of the crypts and the mucosa region (Pradhan et al. ICPRAM 1, 396 [2019]). Figure 1: Translation of physical measurements (Raman spectra and multimodal images) on the left side into biomedical information like diagnostic markers on the right side. [1] Ali et al., Analytical Chemistry 90, 12485 (2018). [2] Houhou et al., Optics Express 28, 21002 (2020). [3] Guo et al., Chemometrics in Raman Spectroscopy, Molecular Sciences and Chemical Engineering, Elsevier (2020). [4] Guo et al., Analytical Chemistry 90, 9787(2018). [5] Pradhan et al., J. of Biophotonics 13, e201960186 (2020). [6] Ryabchykov et al., Frontiers in Chemistry 6, 257 (2018). [7] Bocklitz et al., ICPRAM 1, 874 (2019). [8] Steinbeck et al., Research Ideas and Outcomes 6, e55852 (2020). Contact: Phone: + 49 3641 9-48328 Email: thomas.bocklitz@uni-jena.de
RkJQdWJsaXNoZXIy OTI3Njg=