There is, I think, something ethereal about i —the square root of minus one. I remember first hearing about it at school. It seemed an odd beast at that time—an intruder hovering on the edge of reality.

Usually familiarity dulls this sense of the bizarre, but in the case of i it was the reverse: over the years the sense of its surreal nature intensified. It seemed that it was impossible to write mathematics that described the real world in …

I and i

Probability Distributions.- Linear Models for Regression.- Linear Models for Classification.- Neural Networks.- Kernel Methods.- Sparse Kernel Machines.- Graphical Models.- Mixture Models and EM.- Approximate Inference.- Sampling Methods.- Continuous Latent Variables.- Sequential Data.- Combining Models.

Pattern Recognition and Machine Learning

https://www.zora.uzh.ch/id/eprint/146170/1/1-s2.0-S1078588416301782-main.pdf

Editor's Choice - Management of Descending Thoracic Aorta Diseases: Clinical Practice Guidelines of the European Society for Vascular Surgery (ESVS)

One of the most promising areas of health innovation is the application of artificial intelligence (AI), primarily in medical imaging. This article provides basic definitions of terms such as “machine/deep learning” and analyses the integration of AI into radiology. Publications on AI have drastically increased from about 100–150 per year in 2007–2008 to 700–800 per year in 2016–2017. Magnetic resonance imaging and computed tomography collectively account for more than 50% of current articles. Neuroradiology appears in about one-third of the papers, followed by musculoskeletal, cardiovascular, breast, urogenital, lung/thorax, and abdomen, each representing 6–9% of articles. With an irreversible increase in the amount of data and the possibility to use AI to identify findings either detectable or not by the human eye, radiology is now moving from a subjective perceptual skill to a more objective science. Radiologists, who were on the forefront of the digital era in medicine, can guide the introduction of AI into healthcare. Yet, they will not be replaced because radiology includes communication of diagnosis, consideration of patient’s values and preferences, medical judgment, quality assurance, education, policy-making, and interventional procedures. The higher efficiency provided by AI will allow radiologists to perform more value-added tasks, becoming more visible to patients and playing a vital role in multidisciplinary clinical teams.

/pdf/artificial-intelligence-in-medical-imaging-threat-or-2fzcyw9ypc.pdf

Artificial intelligence in medical imaging: threat or opportunity? Radiologists again at the forefront of innovation in medicine

Embodiments include a system for determining cardiovascular information for a patient. The system may include at least one computer system configured to receive patient-specific data regarding a geometry of the patient's heart, and create a three-dimensional model representing at least a portion of the patient's heart based on the patient-specific data. The at least one computer system may be further configured to create a physics-based model relating to a blood flow characteristic of the patient's heart and determine a fractional flow reserve within the patient's heart based on the three-dimensional model and the physics-based model.

Method and system for patient-specific modeling of blood flow

We performed a thorough evaluation of three main implementation types for outflow boundary conditions of one-dimensional blood flow models: the implicit Lax-Wendroff implementation, the explicit Lax-Wendroff implementation and the implicit method of characteristics implementation. The three implementation types have been applied for resistance and Windkessel outflow boundary conditions. The results show that the implicit Lax-Wendroff method leads to the greatest execution times and can not be used for medium and small vessels due to its divergent behavior. The explicit Lax-Wendroff method offers the best compromise between accuracy and execution speed. The implicit method of characteristics should only be used when the explicit Lax-Wendroff method diverges.

http://ieeexplore.ieee.org/iel5/6138391/6150308/06150403.pdf

Analysis of outflow boundary condition implementations for 1D blood flow models

Graphics Processing Units (GPU) have been used extensively for accelerating parallelizable applications in general, and scientific computations in particular. Stencil based algorithms are used intensively in various research areas and represent good candidates for GPU based acceleration. Since scientific computations have high accuracy requirements, herein we focus on stencil based double precision computations. For a seven-point stencil we introduce two basic implementations, which use two-dimensional and three-dimensional thread organization respectively. Different optimization techniques lead then to a total of seven different implementations, which are evaluated for two NVIDIA Kepler GPUs. The best performance is obtained for the GTX680 card, for a kernel with two-dimensional thread organization and optimized shared memory and register usage.

Double precision stencil computations on Kepler GPUs

Methods We propose a numerical framework for determining pressure field in large arteries, the main components of which are: axi-symmetric 1-D unsteady wave-propagation model with elastic walls and an optimization framework for model personalization via parameter estimation from measured flow and anatomical data (Fig 1). 4D PC MRI data was acquired using a multi-slice phase contrast FLASH cine sequence with 3d velocity encoding (Siemens Magnetom Verio, temporal resolution 52 ms, matrix 100x192, FOV 319x219 mm, slice thickness 5 mm) in a sagittal to coronal slice orientation. Siemens 4D Flow WIP software was used for background phase, motion and venc anti-aliasing corrections, extracting vessel models including centerline trees and lumen boundary, and for automated extraction of flow profiles and cross-sectional areas. We use a linear elastic model for the vessel wall, with an axially varying stiffness. The parameters of the material model are estimated by a numerical optimization procedure, wherein the simulated deformation is fitted to the observed deformation of the vessel wall extracted from the MR data. Unsteady flow rate at the aortic inlet is extracted from PC-MRI data and used as the inlet boundary condition. To obtain physiologically valid mean and pulse pressure values and the wave propagation effects, we use a structured tree outflow boundary condition, wherein a morphological or fractal vessel tree is introduced to model the downstream behavior at the outlets of the vessel tree. The parameters for the outlet boundary condition are then independently perturbed at each individual outlet in order to match the measured unsteady outflow rate from PC-MRI data. This is achieved by adapting the final radius at which the asymmetric binary tree is terminated. The hyperbolic PDEs are solved using finite differences based on 2nd-order Lax-Wendroff method. Simulations converge after 3 cycles, with a maximum relative difference of 1% in pressure and flow between 2 consecutive cycles.

/pdf/determination-of-time-varying-pressure-field-from-phase-2cdag9phkw.pdf

Determination of time-varying pressure field from phase contrast MRI data

Osteoporosis is a skeletal disorder which leads to bone mass loss and to an increased fracture risk. Recently, physics-based models, employing finite element analysis (FEA), have shown great promise in being able to non-invasively estimate biomechanical quantities of interest in the context of osteoporosis. However, these models have high computational demand, limiting their clinical adoption. In this manuscript, we present a deep learning model based on a convolutional neural network (CNN) for predicting average strain as an alternative to physics-based approaches. The model is trained on a large database of synthetically generated cancellous bone anatomies, where the target values are computed using the physics-based FEA model. The performance of the trained model was assessed by comparing the predictions against physics-based computations on a separate test data set. Correlation between deep learning and physics-based predictions was very good (0.895, p < 0.001), and no systematic bias was found in Bland-Altman analysis. The CNN model also performed better than the previously introduced Support Vector Machine (SVM) model which relied on handcrafted features (correlation 0.847, p < 0.001). Compared to the physics based computation, average execution time was reduced by more than 1000 times, leading to real-time assessment of average strain. Average execution time went down from 32.1 ± 3.0 seconds for the FE model to around 0.03 ± 0.005 seconds for the CNN model on a workstation equipped with 3.0 GHz Intel i7 2-core processor.

Towards deep learning based estimation of fracture risk in osteoporosis patients

One of the most active research areas in computed tomography (CT) is to devise a strategy to reduce radiation exposure, while maintaining high image quality, required for accurate diagnosis. The recent advancements offered by deep learning based data-driven approaches for solving inverse problems in biomedical imaging have led to the development of an alternative method for producing high-quality reconstructed images from low-dose CT data. While most of the reconstruction approaches tackle the problem from a post-processing perspective, in this paper, inspired by the idea of unfolding a proximal gradient descent optimization algorithm to finite iterations, and replacing the proximal terms with trainable deep artificial neural networks, we propose an end-to-end solution for reconstructing full-dose tomographic images directly from low-dose measurements. The framework is designed to encapsulate the knowledge of the physical model of CT image formation, and to produce high-quality images that account for human perception through a Generative Adversarial Network with Wasserstein distance and a contextual loss. The proposed method was validated on a clinical dataset, and promising results have been obtained compared to the state-of-the-art mean-squared-error (MSE) based learned iterative reconstruction approach, while also maintaining a runtime suitable for a routine clinical setting.

Lucian Mihai Itu

Papers

Analysis of outflow boundary condition implementations for 1D blood flow models

Double precision stencil computations on Kepler GPUs

Determination of time-varying pressure field from phase contrast MRI data

Towards deep learning based estimation of fracture risk in osteoporosis patients

Data-Driven Adversarial Learning for Sinogram-Based Iterative Low-Dose CT Image Reconstruction