Abstract: To be announced.

Abstract: Over the past years high performance computing for Monte Carlo calculations in radiotherapy has experienced rapid developments, mainly due to the irruption of GPU-based based calculation algorithms. However, the large acceleration factors observed usually come at the cost of loss of generality and absolute accuracy of the simulation. Monte Carlo packages such as EGSnrc have become gold standard due to its extensive testing and comparison with experimental data over the years and therefore it would be desirable that a high performance Monte Carlo algorithm for general purposes retains these characteristics. This work presents OMP_EGS, a novel implementation of parallel processing in the EGSnrc code system using the OpenMP API and optimized for the Intel Many Integrated Core (MIC) architecture. OMP_EGS relies on optimized multi-threading and work- partitioning schemes, together with customized memory allocation strategies to increase the performance gain of the platform. OpenMP directives are easy to implement and can be used for incremental parallelism in a serial application, leveraging the potential of not only MIC co-processors but also modern multi-core architectures such as CPU-based shared memory systems.

Abstract: The recent introduction of integrated light and electron microscopy (ILEM) allows for the extraction of both functional and structural information from a single specimen. The process for acquisition of ILEM data is highly laborious for expert microscopists. Current tools for automated acquisition of electron microscopy (EM) data (see https://youtu.be/h-mT5rxdEZo) do not consider the fact that EM flattens the fluorescence signal in biological samples. A crucial step for an automated process of ILEM is the localisation of cells of interest in fluorescence images, in order to direct further EM imaging of the sample. We propose a simple method that accurately segments and localises most cells in a 2k x 2k image in about 5 seconds, with a user providing just minimum and maximum cell diameter of typical cells and a background-foreground threshold parameter. Our experimental results with images of HeLa cells verify the effectiveness of this method, when compared to the segmentations of expert microscopists. Future work should concern the characterisation of detected cells in order to select the most interesting ones to interrogate them about specific biological questions.

Abstract: The pressure gradient across stenotic blood vessels is an important clinical index for the diagnosis of many cardiovascular diseases. While clinical gold standard for its measurement (in terms of accuracy) is invasive catheterization, Phase-Contrast MR-imaging has emerged as a promising tool for enabling a non-invasive quantification, by linking (highly spatially resolved) velocity measurements with relative pressures via Navier-Stokes equations. In this work we provide a review of current methods for relative pressure estimation, propose new ones and compare them on numerical examples using subsampled and noisy synthetic data. We verify that the newly proposed approaches are more robust with respect to data perturbations and therefore more precise. Joint work with Cristóbal Bertoglio (CMM U. Chile), Rodolfo Núñez (DIM, U. Chile), Felipe Galarce (PUCV) and David Nordsletten (KCL, UK).

Abstract: Level set segmentation has been successfully used in several image applications. However, it performs poorly when applied to severely corrupted images or when they contain poorly defined structures. Shape prior knowledge adds information inferred from a database that allows to compensate the poor data quality. This prior knowledge can be included in level segmentation in two ways: weak prior knowledge, which makes simple geometric assumptions about the shape; and strong prior knowledge, which consists on a regularization term that penalizes shapes that differ from a database. One of the key points of the strong prior knowledge, is that the regularizer needs to be invariant to pose changes of the training shapes, at least under under translation, rotation and scaling. Past research works have accounted for this by coupling the curve evolution to a registration problem through an optimization procedure. This approach is slow and its results depend on how this optimization is implemented. To overcome this issue, Cremers et al. (2006) introduce the intrinsic alignment, which normalizes each shape to a common coordinate system, avoiding the registration process. Nevertheless, their proposed solution considered only scaling and translation but not rotation, which is critical in several image applications. We propose generalization to Cremers' work, that considers intrinsic scaling, translation and rotation. Our regularization term is based on the eigenvalues and eigenvectors of the covariance matrix of each training shape, and this eigendecomposition factors dependency leads into a new set of evolution equations. We tested our regularizer, combined with a Chan-Vese functional, in 2D and 3D synthetic and medical images, demonstrating the effectiveness of using shape priors with intrinsic scaling, translation and rotation alignment in different segmentation problems.

Abstract: In the past few years we have witnessed an explosion in the availability of biomedical data from multiple modalities and scales. However, the lack of automatic methods for interpreting such data presents a major impediment for understanding the mechanisms, diagnosis and treatment of human disease. In this talk, I will overview our recent work on the development of automatic methods for the interpretation of biomedical data from multiple modalities and scales. At the cellular scale, I will present a structured matrix factorization method for segmenting neurons and finding their spiking patterns in calcium imaging videos, and a shape analysis method for classifying embryonic cardiomyocytes in optical imaging videos. At the organ scale, I will present a Riemannian framework for processing diffusion magnetic resonance images of the brain, and a stochastic tracking method for detecting Purkinje fibers in cardiac MRI. At the patient scale, I will present dynamical system and machine learning methods for recognizing surgical gestures and assessing surgeon skill in medical robotic motion and video data.

Abstract: To be announced.

Abstract: The implicit-solvent model uses continuum theory to compute the electrostatic potential around dissolved biomolecules, which yields a system of partial differential equations where the Poisson-Boltzmann and Poisson equations are coupled on the molecular surface. To solve the resulting system of PDEs efficiently, we wrote a fast boundary-element method (with a multipole-based treecode) in Python and CUDA (for exploiting GPUs). We call our code PyGBe — a Python-based GPU code with boundary elements. In this talk, I will detail the implicit-solvent model and our implementation in PyGBe, to then present results in the context of protein solvation, binding, and adsorption, with validation and practical applications. Moreover, despite Poisson-Boltzmann-based solvers are widely used in computational biophysics, the model has well known limitations. Towards the end of this talk, I will discuss the opportunity to overcome such limitations from a boundary integral framework.

Abstract: In the imaging technique of Single-Photon Emission Computed Tomography (SPECT), the transport of photons (gamma-rays) inside the patient can be modelled using the radiative transfer equation (RTE). The image reconstruction problem in SPECT is then described as the reconstruction of the source term in the RTE using the measurement of the RTE solution outside the patient (see e.g. [1]). The standard approach to this reconstruction problem considers only the measurement of the ballistic photons leaving the patient. By assuming that the attenuation coefficient in known, the problem becomes the inversion of the attenuated Radon transform (see e.g. [3]). While the problem of reconstructing both, the source and the attenuation coefficients, from the ballistic measurements is commonly referred to as the identification problem (see e.g. [4]). In this talk I will present an approach in which we try to tackle the problem by assuming an enlarged set of measurements, namely the ballistic and first order scattering photons. By considering the decomposition of the RTE in orders of scattering, in [2] we proposed a model equation for these measurements and studied the inversion of the linearized equation for small attenuations, obtaining also some partial results for the original non-linear equation and some numerical validation with synthetic experimentation. I will present these results and also I will present some ideas we are exploring to bring this mathematical results closer to the SPECT application. Joint work with J. C. Quintana, F. Monard, A. Osses, F. Remero, and CIB–UC.

References:
[1] G. Bal. Inverse transport theory and applications, Inv. Prob., 25, 2009.
[2] M. Courdurier, F. Monard, A. Osses, F. Romero. Simultaneous source and attenuation reconstruction in SPECT using ballistic and single scattering data, Inv. Prob., 31, (2015).
[3] R. Novikov. An inversion formula for the attenuated x-ray transformation, Arkiv för Matematik, 40, 145– 67, 2002.
[4] P. Stefanov. The Identification Problem for the attenuated X-ray transform, Amer. J. Math., 136(5): 1215– 1247, 2014.

Abstract: Inverse Problems arise whenever the parameters of a model have to be inferred from observed data. They are almost always ill-posed and sometimes non-linear. In this talk I will give an overview of different inverse problems in medical imaging for different physical models and observations including optical, acoustic, and electromagnetic data. I will stress the role of computational modelling and regularisation. I will illustrate with examples from projects at the UCL Centre for Medical Computing (CMIC).

Abstract: To be announced.

Abstract: To be announced.

Abstract: To be announced.

Abstract: To be announced.

Abstract: To be announced.