Search Results
FLIM data analysis based on Laguerre polynomial decomposition and machine-learning
2021, Guo, Shuxia, Silge, Anja, Bae, Hyeonsoo, Tolstik, Tatiana, Meyer, Tobias, Matziolis, Georg, Schmitt, Michael, Popp, Jürgen, Bocklitz, Thomas
Significance: The potential of fluorescence lifetime imaging microscopy (FLIM) is recently being recognized, especially in biological studies. However, FLIM does not directly measure the lifetimes, rather it records the fluorescence decay traces. The lifetimes and/or abundances have to be estimated from these traces during the phase of data processing. To precisely estimate these parameters is challenging and requires a well-designed computer program. Conventionally employed methods, which are based on curve fitting, are computationally expensive and limited in performance especially for highly noisy FLIM data. The graphical analysis, while free of fit, requires calibration samples for a quantitative analysis. Aim: We propose to extract the lifetimes and abundances directly from the decay traces through machine learning (ML). Approach: The ML-based approach was verified with simulated testing data in which the lifetimes and abundances were known exactly. Thereafter, we compared its performance with the commercial software SPCImage based on datasets measured from biological samples on a time-correlated single photon counting system. We reconstructed the decay traces using the lifetime and abundance values estimated by ML and SPCImage methods and utilized the root-mean-squared-error (RMSE) as marker. Results: The RMSE, which represents the difference between the reconstructed and measured decay traces, was observed to be lower for ML than for SPCImage. In addition, we could demonstrate with a three-component analysis the high potential and flexibility of the ML method to deal with more than two lifetime components.
Computational tissue staining of non-linear multimodal imaging using supervised and unsupervised deep learning
2021, Pradhan, Pranita, Meyer, Tobias, Vieth, Michael, Stallmach, Andreas, Waldner, Maximilian, Schmitt, Michael, Popp, Juergen, Bocklitz, Thomas
Hematoxylin and Eosin (H&E) staining is the 'gold-standard' method in histopathology. However, standard H&E staining of high-quality tissue sections requires long sample preparation times including sample embedding, which restricts its application for 'real-time' disease diagnosis. Due to this reason, a label-free alternative technique like non-linear multimodal (NLM) imaging, which is the combination of three non-linear optical modalities including coherent anti-Stokes Raman scattering, two-photon excitation fluorescence and second-harmonic generation, is proposed in this work. To correlate the information of the NLM images with H&E images, this work proposes computational staining of NLM images using deep learning models in a supervised and an unsupervised approach. In the supervised and the unsupervised approach, conditional generative adversarial networks (CGANs) and cycle conditional generative adversarial networks (cycle CGANs) are used, respectively. Both CGAN and cycle CGAN models generate pseudo H&E images, which are quantitatively analyzed based on mean squared error, structure similarity index and color shading similarity index. The mean of the three metrics calculated for the computationally generated H&E images indicate significant performance. Thus, utilizing CGAN and cycle CGAN models for computational staining is beneficial for diagnostic applications without performing a laboratory-based staining procedure. To the author's best knowledge, it is the first time that NLM images are computationally stained to H&E images using GANs in an unsupervised manner.
Neural partial differential equations for chaotic systems
2021, Gelbrecht, Maximilian, Boers, Niklas, Kurths, Jürgen
When predicting complex systems one typically relies on differential equation which can often be incomplete, missing unknown influences or higher order effects. By augmenting the equations with artificial neural networks we can compensate these deficiencies. We show that this can be used to predict paradigmatic, high-dimensional chaotic partial differential equations even when only short and incomplete datasets are available. The forecast horizon for these high dimensional systems is about an order of magnitude larger than the length of the training data.