CartesianToPolar and PolarToCartesian Image Filters

In this paper, we describe a set of filters, implemented in the Insight Toolkit www.itk.org, for converting an image from Cartesian co-ordinate space to Polar co-ordinate space and vice-versa. Cartesian to Polar conversion of an image is a useful operation in preprocessing stage of certain image-processing algorithm where feature of interest has simplified representation in the polar space. This paper is accompanied with the source code, input data, parameters and output data that the authors used for validating the algorithm described in this paper. This adheres to the fundamental principle that scientific publications must facilitate reproducibility of the reported results.

An Empirical Optimization to Logistic Classification Model

Recently, the scientific community has been proposing several automatic algorithms to biomedical image segmentation procedure, being an interesting and helpful approach to assist both technicians and radiologists in this time-consuming and subjective task. One of these interesting and widely used image segmentation method could be the voxel intensity-based algorithms, e.g. image histogram threshold methods, which have been intensively improved in the past decades. Recently, an interesting approach that gained focus is the logistic classification (LC) for object detection in biomedical images. Even though the general concept behind the LC method is fairly known, the proper method’s optimization still commonly adjusted by hand which naturally adds a level of uncertainty and subjectivity in the general segmentation performance. Therefore, an empirical LC optimization is presented, offering a ITK class that performs the LC parameters optimization based on empirical input data analysis. It is worth mentioning that the LogisticContrastEnhancementImageFilter class showed here is also applied on others computational problems, being briefly explained in this document.

Adaptive Moment Estimator (Adam) Optimizer in ITK v3

This document describes an ITK class implementing an Adaptive Moment Estimator (Adam) optimizer algorithm within the Insight Toolkit ITK www.itk.org. Adam is an adaptive gradient descent optimizer, which independently adaptively estimates the gradient descent step for each parameter, at each iteration, based on stored past gradients. The optimizer stores exponentially decaying averages of past gradients to estimate first moment (the mean) and the second moment (the variance) of the gradients to formulate update rule for present iteration. The Adam optimizer compares favorably to other adaptive learning-method algorithms, converges faster, and is robust to saddle point. This paper is accompanied with the source code, input data, parameters and output data that the authors used for validating the algorithm described in this paper.

Scalable Simple Linear Iterative Clustering (SSLIC) Using a Generic and Parallel Approach

Superpixel algorithms have proven to be a useful initial step for segmentation and subsequent processing of images, reducing computational complexity by replacing the use of expensive per-pixel primitives with a higher-level abstraction, superpixels. They have been successfully applied both in the context of traditional image analysis and deep learning based approaches. In this work, we present a general- ized implementation of the simple linear iterative clustering (SLIC) superpixel algorithm that has been generalized for n-dimensional scalar and multi-channel images. Additionally, the standard iterative im- plementation is replaced by a parallel, multi-threaded one. We describe the implementation details and analyze its scalability using a strong scaling formulation. Quantitative evaluation is performed using a 3D image, the Visible Human cryosection dataset, and a 2D image from the same dataset. Results show good scalability with runtime gains even when using a large number of threads that exceeds the physical number of available cores (hyperthreading).

Computing Bone Morphometric Feature Maps from 3-Dimensional Images

This document describes a new remote module implemented for the Insight Toolkit (ITK), itkBoneMorphometry. This module contains bone analysis filters that compute features from N-dimensional images that represent the internal architecture of bone. The computation of the bone morphometry features in this module is based on well known methods. The two filters contained in this module are itkBoneMorphometryFeaturesFilter. which computes a set of features that describe the whole input image in the form of a feature vector, and itkBoneMorphometryFeaturesImageFilter, which computes an N-D feature map that locally describes the input image (i.e. for every voxel). itkBoneMorphometryFeaturesImageFilter can be configured based in the locality of the desired morphometry features by specifying the neighborhood size. This paper is accompanied by the source code, the input data, the choice of parameters and the output data that we have used for validating the algorithms described. This adheres to the fundamental principle that scientific publications must facilitate reproducibility of the reported results.

Efficient multithreading for manycore processor: Multidimensional domain decomposition using Intel® TBB

The Insight Toolkit (ITK) utilizes a generic design for image processing filters that allows many developers to rapidly implement new algorithms. While ITK filters benefit from a platform-independent and versatile multithreading capability, the current implementation does not easily achieve high performance. First, ITK relies on a static decomposition of the image into subsets of equal size which is highly inefficient when the computational complexity varies between subsets (unbalanced workloads). Second, the current domain decomposition is limited to subdivide the input domain along a single dimension (typically the slice dimension in a 3-D volume), which causes a multithreading under-utilization when the number of threads is larger than the size of this dimension when using massively parallel compute systems. We previously presented a new itk::TBBImageToImageFilter class that replaced the static task decomposition by a dynamic task decomposition for improved workload balancing, in which the job scheduling task was optimized using the Intel® Threading Building Blocks (TBB) library. In this work, we propose a new multidimensional dynamic image decomposition approach that allows decomposition over an arbitrary number of dimensions. This new generic multithreading capability, combined with the TBB dynamic task scheduler, substantially improves multithreading performance when using massively parallel processors.

Computing Textural Feature Maps for N-Dimensional images

This document describes a new remote module implemented for the Insight Toolkit ITK, itkTextureFeatures. This module contains two texture analysis filters that are used to compute feature maps of N-Dimensional images using two well-known texture analysis methods. The two filters contained in this module are itkScalarImageToTextureFeaturesImageFilter (which computes textural features based on intensity-based co-occurrence matrices in the image) and itkScalarImageToRunLengthFeaturesImageFilter (which computes textural features based on equally valued intensity clusters of different sizes or run lengths in the image). The output of this module is a vector image of the same size than the input that contains a multidimensional vector in each pixel/voxel. Filters can be configured based in the locality of the textural features (neighborhood size), offset directions for co-ocurrence and run length computation, the number of bins for the intensity histograms, the intensity range or the range of run lengths. This paper is accompanied with the source code, input data, parameters and output data that we have used for validating the algorithm described in this paper. This adheres to the fundamental principle that scientific publications must facilitate reproducibility of the reported results.

N-Dimensional Computation of Strain Tensor Images in the Insight Toolkit

Strain quantifies local deformation of a solid body. In medical imaging, strain reflects how tissue deforms under load. Or, it can quantify growth or atrophy of tissue, such as the growth of a tumor. Additionally, strain from the transformation that results from image-to-image registration can be applied as an input to a biomechanical constitutive model.This document describes N-dimensional computation of strain tensor images in the Insight Toolkit (ITK), www.itk.org. Two filters are described. The first filter computes a strain tensor image from a displacement field image. The second filter computes a strain tensor image from a general spatial transform. In both cases, infinitesimal, Green-Lagrangian, or Eulerian-Almansi strain can be generated.This paper is accompanied with the source code, input data, parameters and output data that the authors used for validating the algorithm described in this paper. This adheres to the fundamental principle that scientific publications must facilitate reproducibility of the reported results.

Automatic Conductance Estimation Methods for Anisotropic Diffusion ITK Filters

The anisotropic diffusion algorithm has been intensively studied in the past decades, which could be considered as a very efficient image denoising procedure in many biomedical applications. Several authors contributed many clever solutions for diffusion parameters fitting in specific imaging modalities. Furthermore, besides improvements regarding the image denoising quality, one important variable that must be carefully set is the conductance, which regulates the structural edges preservation among the objects presented in the image. The conductance value is strongly dependent on image noise level and an appropriate parameter setting is, usually, difficult to find for different images databases and modalities. Fortunately, thanks to many efforts from the scientific community, a few automatic methods have been proposed in order to set the conductance value automatically. Here, it is presented an ITK class which offers a simple collection of the most common automatic conductance setting approaches in order to assist researchers in image denoising procedures using anisotropic-based filtering methods (such as well described in the AnisotropicDiffusionFunction class).

A patch-based framework for new ITK functionality: Joint fusion, denoising, and non-local super-resolution

In an earlier Insight Journal article, we introduced an ITK implementation of the adaptive patch-based image denoising algorithm described in [3]. We follow-up up that offering with a generalized non-local, patch-based ITK class framework and a refactored denoising class. In addition, we provide two ITK implementations of related, well-known algorithms. The first is a non-local super resolution method described in [1, 2]. The second is the multivariate joint label fusion algorithm of [4, 5] with additional extensions, denoted as “joint intensity fusion”, which will be described in a forthcoming manuscript. Accompanying these ITK classes are documented programming interfaces which use our previously introduced unique command line interface routines. Several 2-D examples on brain imaging data are provided to qualitatively demonstrate performance.