2018

  • Abstract:

    Recent breakthroughs in magnetic resonance imaging (MRI) enabled quantitative relaxometry and diffusion-weighted imaging with sub-millimeter resolution. Combined with biophysical models of MR contrast the emerging methods promise in vivo mapping of cyto- and myelo-architectonics, i.e., in vivo histology using MRI (hMRI) in humans. The hMRI methods require histological reference data for model building and validation. This is currently provided by MRI on post mortem human brain tissue in combination with classical histology on sections. However, this well established approach is limited to qualitative 2D information, while a systematic validation of hMRI requires quantitative 3D information on macroscopic voxels. We present a promising histological method based on optical 3D imaging combined with a tissue clearing method, Clear Lipid-exchanged Acrylamide-hybridized Rigid Imaging compatible Tissue hYdrogel (CLARITY), adapted for hMRI validation. Adapting CLARITY to the needs of hMRI is challenging due to poor antibody penetration into large sample volumes and high opacity of aged post mortem human brain tissue. In a pilot experiment we achieved transparency of up to 8 mm-thick and immunohistochemical staining of up to 5 mm-thick post mortem brain tissue by a combination of active and passive clearing, prolonged clearing and staining times. We combined 3D optical imaging of the cleared samples with tailored image processing methods. We demonstrated the feasibility for quantification of neuron density, fiber orientation distribution and cell type classification within a volume with size similar to a typical MRI voxel. The presented combination of MRI, 3D optical microscopy and image processing is a promising tool for validation of MRI-based microstructure estimates.

    PubMed

2019

  • Jan-Matthis Lueckmann, Giacomo Bassetto, Theofanis Karaletsos, Jakob H. Macke ; 
    Proceedings of The 1st Symposium on Advances in Approximate Bayesian Inference, PMLR 96:32-53, 2019.
     
     

    Abstract:

    Approximate Bayesian Computation (ABC) provides methods for Bayesian inference in simulation-based models which do not permit tractable likelihoods. We present a new ABC method which uses probabilistic neural emulator networks to learn synthetic likelihoods on simulated data – both ’local’ emulators which approximate the likelihood for specific observed data, as well as ’global’ ones which are applicable to a range of data. Simulations are chosen adaptively using an acquisition function which takes into account uncertainty about either the posterior distribution of interest, or the parameters of the emulator. Our approach does not rely on user-defined rejection thresholds or distance functions. We illustrate inference with emulator networks on synthetic examples and on a biophysical neuron model, and show that emulators allow accurate and efficient inference even on problems which are challenging for conventional ABC approaches.

    PMLR

  • David Greenberg, Marcel Nonnenmacher, Jakob Macke ; 
    Proceedings of the 36th International Conference on Machine Learning, PMLR 97:2404-2414, 2019.
     
     

    Abstract:

    How can one perform Bayesian inference on stochastic simulators with intractable likelihoods? A recent approach is to learn the posterior from adaptively proposed simulations using neural network-based conditional density estimators. However, existing methods are limited to a narrow range of proposal distributions or require importance weighting that can limit performance in practice. Here we present automatic posterior transformation (APT), a new sequential neural posterior estimation method for simulation-based inference. APT can modify the posterior estimate using arbitrary, dynamically updated proposals, and is compatible with powerful flow-based density estimators. It is more flexible, scalable and efficient than previous simulation-based inference techniques. APT can operate directly on high-dimensional time series and image data, opening up new applications for likelihood-free inference.

    PMLR

  •  

    Jana Bürgers, Irina Pavlova, Juan E. Rodriguez-Gatica, Christian Henneberger, Marc Oeller, Jan A. Ruland, Jan P. Siebrasse, Ulrich Kubitscheck, Martin K. Schwarz; Neurophotonics 6(1), 015005 (2019), doi: 10.1117/1.NPh.6.1.015005.

     
     

    Abstract:

    The goal of understanding the architecture of neural circuits at the synapse level with a brain-wide perspective has powered the interest in high-speed and large field-of view volumetric imaging at subcellular resolution. Here we developed a method combining tissue expansion and light sheet fluorescence microscopy to allow extended volumetric super resolution high-speed imaging of large mouse brain samples. We demonstrate the capabilities of this method by performing two color fast volumetric super resolution imaging of mouse CA1 and dentate gyrus molecular-, granule cell- and polymorphic layers. Our method enables an exact evaluation of granule cell and neurite morphology within the context of large cell ensembles spanning several orders of magnitude in resolution. We found that imaging a brain region of 1 mm3 in super resolution using light sheet fluorescence expansion microscopy is about 17-fold faster than imaging the same region by a current state of the art high resolution confocal laser scanning microscope.

    PMLR