Visualizing the uncertainty of ensemble simulations is challenging
due to the large size and multivariate and temporal features of ensemble data sets. One popular approach to studying the uncertainty of ensembles is analyzing the positional uncertainty of the level sets. Probabilistic marching cubes is a technique that performs Monte
Carlo sampling of multivariate Gaussian noise distributions for positional uncertainty visualization of level sets. However, the technique suffers from high computational time, making interactive visualization and analysis impossible to achieve. This paper introduces a
deep-learning-based approach to learning the level-set uncertainty for two-dimensional ensemble data with a multivariate Gaussian noise assumption. We train the model using the first few time steps from time-varying ensemble data in our workflow. We demonstrate
that our trained model accurately infers uncertainty in level sets for new time steps and is up to 170X faster than that of the original probabilistic model with serial computation and 10X faster than that of the original parallel computation.
due to the large size and multivariate and temporal features of ensemble data sets. One popular approach to studying the uncertainty of ensembles is analyzing the positional uncertainty of the level sets. Probabilistic marching cubes is a technique that performs Monte
Carlo sampling of multivariate Gaussian noise distributions for positional uncertainty visualization of level sets. However, the technique suffers from high computational time, making interactive visualization and analysis impossible to achieve. This paper introduces a
deep-learning-based approach to learning the level-set uncertainty for two-dimensional ensemble data with a multivariate Gaussian noise assumption. We train the model using the first few time steps from time-varying ensemble data in our workflow. We demonstrate
that our trained model accurately infers uncertainty in level sets for new time steps and is up to 170X faster than that of the original probabilistic model with serial computation and 10X faster than that of the original parallel computation.
Machine Learning & Information Processing, Proceedings of ICMLIP 2020
Editors: D. Swain, P. K. Pattnaik, T. M. Athawale
[Book link]Â
Abstract
This book includes selected papers from the 2nd International Conference on Machine Learning and Information Processing (ICMLIP 2020), held at Vardhaman College of Engineering, Jawaharlal Nehru Technological University (JNTU), Hyderabad, India, from November 28 to 29, 2020. It presents the latest developments and technical solutions in the areas of advanced computing and data sciences, covering machine learning, artificial intelligence, human–computer interaction, IoT, deep learning, image processing and pattern recognition, and signal and speech processing.
Deep brain stimulation (DBS) is an FDA-approved neurosurgical procedure for treating patients with movement disorders such as Parkinson's disease. Patient-specific computational modeling and visualization play a key role for efficient surgical and therapeutic decision-making relevant to DBS. The computational models analyze DBS post-operative brain imaging, e.g., computed tomography (CT), to understand the DBS electrode positions within the patient's brain. The DBS stimulation settings for optimal patient response depend upon a physician's knowledge regarding precise electrode positions. The finite resolution of brain imaging, however, restricts our understanding regarding precise DBS electrode positions. In our contribution, we study the problem of the quantification of positional uncertainty in the DBS electrodes caused by the finite resolution of post-operative imaging. We propose a Monte Carlo statistical framework, which takes the advantage of our analytical characterization of the DBS electrode geometry to understand the spatial uncertainty in DBS electrodes. Our statistical framework quantifies the uncertainty in two positional parameters of the DBS electrodes, namely, the longitudinal axis direction and the positions at sub-voxel levels. We interactively visualize quantified uncertainties by employing volume rendering and isosurfaces. We show that the spatial variations in the DBS electrode positions are significant for finite resolution imaging, and interactive visualization can be instrumental for efficient interpretation of the positional variations in the DBS lead.