Information Theoretic Methods for Diffusion-Weighted MRI Analysis

Multioutput regression has recently shown great ability to solve challenging problems in both computer vision and medical image analysis. However, due to the huge image variability and ambiguity, it is fundamentally challenging to handle the highly complex input-target relationship of multioutput regression, especially with indiscriminate high-dimensional representations. In this paper, we propose a novel supervised descriptor learning (SDL) algorithm for multioutput regression, which can establish discriminative and compact feature representations to improve the multivariate estimation performance. The SDL is formulated as generalized low-rank approximations of matrices with a supervised manifold regularization. The SDL is able to simultaneously extract discriminative features closely related to multivariate targets and remove irrelevant and redundant information by transforming raw features into a new low-dimensional space aligned to targets. The achieved discriminative while compact descriptor largely reduces the variability and ambiguity for multioutput regression, which enables more accurate and efficient multivariate estimation. We conduct extensive evaluation of the proposed SDL on both synthetic data and real-world multioutput regression tasks for both computer vision and medical image analysis. Experimental results have shown that the proposed SDL can achieve high multivariate estimation accuracy on all tasks and largely outperforms the algorithms in the state of the arts. Our method establishes a novel SDL framework for multioutput regression, which can be widely used to boost the performance in different applications.

In image processing and computer vision applications such as medical or scientific image data analysis, as well as in industrial scenarios, images are used as input measurement data. It is good scientific practice that proper measurements must be equipped with error and uncertainty estimates. For many applications, not only the measured values but also their errors and uncertainties, should be—and more and more frequently are—taken into account for further processing. This error and uncertainty propagation must be done for every processing step such that the final result comes with a reliable precision estimate. The goal of this book is to introduce the reader to the recent advances from the field of uncertainty quantification and error propagation for computer vision, image processing, and image analysis that are based on partial differential equations (PDEs). It presents a concept with which error propagation and sensitivity analysis can be formulated with a set of basic operations. The approac...

The goal of this book was to introduce the reader to the recent advances in the field of uncertainty quantification and error propagation for computer vision, image processing, and image analysis that are based on partial differential equations (PDEs). Our original motivation was the observation that more and more often, measured data in the form of images are being used as part of the simulation pipeline, and that error propagation and uncertainty quantification—starting from images but then moving through the pipeline—are important to understanding the ending result of this scientific or engineering process. To analyze the image processing pipeline, we have presented concepts that enable error propagation to be formulated with a set of basic operations: the idea of stochastic images and corresponding (numerical PDE) operations on those images. From this perspective, one approach commonly used within the engineering literature is to employ Monte Carlo sampling techniques to attempt to quantify the impact of errors and variability on an engineering pipeline. Within the uncertainty quantification world, alternative approaches such as the class of generalized polynomial chaos (gPC) methods, which make assumptions about the ability to approximate the process of interest through polynomial approximations, have gained traction as a way of accelerating the convergence of the error and uncertainty quantification process in a computationally tractable way. In this book, we have shown the results of our exploration of the use of the gPC methodology for image processing and computer vision problems. We have relied on the fact that all the prerequisites of the gPC framework (e.g., finite variance, smoothness in the stochastic space, etc.) are intrinsically satisfied in the image processing context. In particular, we have shown how the generalized polynomial chaos (gPC) approach and its corresponding rules for computation (e.g., sums, products, projections on the gPC space, etc.) lead to straightforward generalizations to image processing and computer vision techniques. We have shown that the gPC methodology, when combined with image processing and computer vision techniques, provides a powerful approach to error propagation and uncertainty quantification, and to accessing system sensitivity.

Improving the quality of medical images at pre- and post-surgery operations are necessary for beginning and speeding up the recovery process. Partial differential equations-based models have become a powerful and well-known tool in different areas of image processing such as denoising, multiscale image analysis, edge detection and other fields of image processing and computer vision. In this paper, an algorithm for medical image denoising using anisotropic diffusion filter with a convenient stopping criterion is presented. In this regard, the current paper introduces two strategies: utilizing the efficient explicit method due to its advantages with presenting impressive software technique to effectively solve the anisotropic diffusion filter which is mathematically unstable, proposing an automatic stopping criterion, that takes into consideration just input image, as opposed to other stopping criteria, besides the quality of denoised image, easiness and time. Various medical images are examined to confirm the claim.

Deep Learning (DL) revolutionized a wide variety of research fields and applications of Artificial Intelligence (AI) in the last years. Computer vision is among those fields that witnessed a stronger impact, especially in some tasks like image equalization, classification and segmentation, that have several applications in medical imaging. Such applications often focus on microscopy images, but they can be successfully extended to other fields, like the detection of diseases involving the ocular surface and the cornea.One of the reasons of the success of deep learning applied to image processing relies in the end‐to‐end data‐driven approach. This means that the whole image processing and classification/segmentation pipeline can be adapted to the data being processed. This is particularly powerful when dealing with medical images, whose processing is often based on the detection of low‐ and mid‐level patterns that might be difficult to define in a formal manner. This means that deep learning has the chance to select and consider some image elements that would be nearly impossible to define using standard computer vision techniques. However, this becomes possible if large amounts of data are available for training the network, which became reality in the last years thanks to the wide digitalization of medical systems, including ophthalmology equipment.Based on a substantial amount of literature that has rapidly grown in the last years, this talk will focus on the new perspectives of eye disease diagnosis that were opened by the advent of deep learning. At the same time, the limitations and risks that must be considered in the development of DL‐based medical systems, like overfitting and lack of generalization, will be highlighted and discussed. Keeping track of such limitations is crucial in order not to consider DL as a sort of magic box, but rather, as powerful image analysis and processing tool.The talk will also provide a qualitative and quantitative evaluation of the diagnosis systems currently available that exploit DL and AI systems. The current and future impact of automatic image processing systems dedicated to ophthalmology will also be discussed. When automatic medical image analysis will be widely available, the patterns of medical diagnosis will need to change, taking advantage of the possibility to perform larger screenings that will be automatically processed.

This article introduces a medical imaging and image processing knowledge center which serves as an e-learning platform based on network technology. This platform includes contents of medical imaging and medical image processing for the teaching activities, virtual reality tutoring system of human skeleton, medical image database, etc. The purpose of establishing this platform is to recruit and excavate information records which usually associated with the daily data exchanged in teaching and research activities. Then it will gradually form an open professional knowledge database. At present, this platform plays the role of knowledge community in the educational activities of BME program in Tsinghua University.

Symmetric positive-definite (SPD) matrices are ubiquitous in Computer Vision, Machine Learning and Medical Image Analysis. Finding the center/average of a population of such matrices is a common theme in many algorithms such as clustering, segmentation, principal geodesic analysis, etc. The center of a population of such matrices can be defined using a variety of distance/divergence measures as the minimizer of the sum of squared distances/divergences from the unknown center to the members of the population. It is well known that the computation of the Karcher mean for the space of SPD matrices which is a negatively-curved Riemannian manifold is computationally expensive. Recently, the LogDet divergence-based center was shown to be a computationally attractive alternative. However, the LogDet-based mean of more than two matrices can not be computed in closed form, which makes it computationally less attractive for large populations. In this paper we present a novel recursive estimator for center based on the Stein distance - which is the square root of the LogDet divergence - that is significantly faster than the batch mode computation of this center. The key theoretical contribution is a closed-form solution for the weighted Stein center of two SPD matrices, which is used in the recursive computation of the Stein center for a population of SPD matrices. Additionally, we show experimental evidence of the convergence of our recursive Stein center estimator to the batch mode Stein center. We present applications of our recursive estimator to K-means clustering and image indexing depicting significant time gains over corresponding algorithms that use the batch mode computations. For the latter application, we develop novel hashing functions using the Stein distance and apply it to publicly available data sets, and experimental results have shown favorable comparisons to other competing methods.

Active contour model (ACM) plays an important role in computer vision and medical image analysis. The traditional ACMs were employed to extract closed contours of objects. While simultaneous extraction of line- and block-like objects, such as boundary contours, centerlines, as well as their topological relationship, remains open so far. Therefore, a novel ACM named "Ingenious Snake" is proposed to adaptively extract the feature curves. The proposed ingenious snake includes the following steps: 1) In the preprocessing, the line- and block-like structures are classified with k-means clustering, following up with morphological operation and enhancement. The gradient vector flow (GVF) field is then acquired from the resultant ridge feature map. 2) For the automatic initialization, the ridge-points are extracted by using the local phase measurement of GVF field, then the two-category of object ridgelines are obtained fast. 3) Finally, the contour deformation and curves evolvement are implemented with a management strategy. The resultant contours and centerlines well characterize the objects of interest. In the experiments, we compare the existing initialization methods and the adaptive extraction with a series of phantoms and testing images. The visual and quantitative assessments of structure extraction are satisfying in terms of effectiveness and accuracy.

Classification and recognition of objects for images is an important issue in many scientific fields such as computer vision, biometrics and medical image analysis. An important feature of many objects is shape, so shape analysis has become an important part of classification. One method of shape analysis is to estimate boundaries and analyze the shape of these boundaries to determining the characteristics of the original object. However, many literature studies on point cloud shape analysis are based on existing shapes. This paper mainly refers to Ripley’s K-function in spatial point analysis, through this judgment on spatial distribution of point cloud data to determine the existence of shape in point cloud data, through the spatial distribution of 2D point cloud data and 3D point cloud data. Judging by the randomness of the experiment, K-function has a considerable effect on judging existence of point cloud data shape through relevant experimental verification analysis.

Statistical models of the shape and appearance of deformable objects have become widely used in Computer Vision and Medical Image Analysis. Here we give an overview of such models and of two efficient algorithms for matching such models to new images (Active Shape Models and Active Appearance Models). We also describe recent work on automatically constructing such models from minimally labelled training images.KeywordsMinimum Description LengthFace ModelActive Appearance ModelActive Shape ModelMedical Image AnalysisThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.

Ground truth or reference data is the basis for performance analysis in computer vision and image processing. This paper summarizes design considerations for ground truth acquisition with a focus on accuracy and cost of different approaches. I argue that first and foremost, requirements engineering needs to be carried out and it needs to be decided whether quantitative performance analysis is beneficial to the field of research at all. Three categories of reference data are presented. Examples for each approach are given and advantages and downsides discussed. Finally, open issues are briefly summarized.

Monitoring cameras are now widely used to monitor everything from a room in a house to an entire warehouse. However, in real monitoring scenarios, a variety of factors, such as underexposure, optical blurring, defocusing, have an impact on the quality of images, which leads to low-quality and low-resolution (LR) of the individual of interest. Reconstruction of a high-resolution (HR) face image with detailed facial features, from a LR observation based on a set of HR and LR training image pairs, plays an important role in computer vision and face image analysis applications. To super-resolve an HR face given a LR face image, the key issue is how to effectively encode the LR image patch. However, due to stability and accuracy issues, the coding approaches proposed so far are far from satisfactory. In this paper, we present a novel sparse coding method via exploiting the support information on the coding coefficients. According to the distances between the input patch and bases in the dictionary, we first assign different weights to the coding coefficients and then obtain the coding coefficients by solving a weighted sparse problem. Experiments on commonly used databases and some face images on the real monitoring conditions demonstrate that our method outperforms state-of-the-art.

Assessing sensing techniques for detecting markers of approaching ecdysis in juvenile tropical rock lobsters, Panulirus ornatus

Polygonal approximation (PA) is a challenging problem in representation of images in computer vision, pattern recognition and image analysis. This paper proposes a stochastic technique-based firefly algorithm (FA) for PA. This technique customizes a kind of randomization by searching a set of solutions. In contrast, PA requires more combination of approximation to find an optimal solution. The algorithm involves several steps to produce better results. The attractiveness and brightness of the firefly have been used efficiently to solve the approximation problem. While compared to other similar algorithms, FA is independent of velocities which are considered as an advantage for this algorithm. Subsequently, the multi-swarm nature of FA allows finding multiple optimal solutions concurrently. This technique achieves the main goal of PA that is minimum error value with less number of dominant points. The experimental results show that proposed algorithm generates better solutions compared to other algorithms.KeywordsPolygonal approximationFirefly algorithmError valueDominant pointMulti-swarm natureVelocitiesOptimal solutions

Single-molecule nanocircuits based on field-effect transistors (smFETs) have known a rapid development and promising results for the functional detection of biomolecular structures and dynamics at the single-molecule scale [1]. In fact, thanks to the size compatibility between the target analyte and the transducer, most often a carbon nanotube, this label-free and amplification-free single-molecule sensing technique allows real-time monitoring of the rapid transitions between different biochemical conformational or interaction states, such as hybridization or folding in nucleic acids. The stability of smFET signals also enables long acquisition periods of such single-molecule interactions with high throughput, allowing to reveal individual reaction pathways usually hidden in average measurements of ensemble methods, and to record time trajectories containing rare or short-lived biomolecular events. Detecting and modeling the kinetics and thermodynamics of biochemical events from smFET recordings requires robust data analysis tools that can idealize these signals into discrete state trajectories, corresponding to successive biochemical states and transitions between them. However, most of the available single-molecule data analysis techniques have been developed for fluorescence and force-based single-molecule experiments, which differ significantly from FET-based experiments, and are thus difficult to adapt to smFET signals. Analysis of smFET time series requires to handle the following set of challenging signal specificities: 1) the stochastic nature of the biomolecular system, 2) the possible non-stationarities in the molecular dynamics of the reaction system, such as changes between transient and steady-state conformations, 3) the multi-source composition of the sensor response, aggregating all contributions from the biochemical system with those from the environment and sensor components in a single input, 4) the mixed noises (AWGN, flicker, and impulse) characteristic of FET devices, 5) the slow baseline drift observed in long acquisitions, and 6) the sizable amount of data generated by such recordings. Here, we propose a new approach for smFET data idealization based on information theory and machine learning for signal processing. We present computational methods designed to achieve automatic detection of molecular events without prior knowledge on the data generating process or signal pre-filtering, which are especially tailored for large, drifting and non-stationary time series that are typical of smFET experiments. First, we address the problem of compensating slow baseline drift, due to sensor degradation or variations in environmental parameters. Such drift can introduce systematic errors in the conversion of signal into discrete states. We developed a 3-step adaptive blind source separation algorithm to decompose multicomponent signals into their embedded layers, thus allowing to separate the contribution of the drift from those of signal and noise [2]. To do so, our algorithm is composed of the following steps: 1) iterative multiscale signal compression based on a minimum description length objective function, 2) unsupervised dynamic drift node positioning, and 3) adaptive piecewise cubic Hermite interpolation. Second, we address the task of trace idealization as a piecewise regression fit of the recorded signal, in order to extract the different biochemical states and rates and thus learn kinetic parameters. To this end, we propose a 4-step hybrid algorithm called cc-EM (compressed clustered Expectation-Maximization): 1) A model-free and unsupervised compression stage is applied to transform the raw signal into a piecewise trajectory. The compression is based on a minimum description length objective function that encodes the entropy of the stochastic sources to achieve minimum redundancies for maximum relevance [3], [4]. 2) A clustering step, based on k-medoids with a swapping cost function, is then applied on compression patterns to gather, without supervision, similar sub-states into common parent states. In traces with concept drift, such soft-clustering enables to reach the lower bound of entropy, corresponding to the optimal compression ratio and learning rate, which drastically decreases the false positive event detection rate and the risk of overfitting. 3) An expectation-maximization (EM) refinement is added to infer possible missed states and to correct the location of transitions to obtain a more accurate idealized trace. 4) Finally, a model selection algorithm screens the compressed-clustered states space domain to select the best fitting model. The proposed algorithms were tested on simulated smFET signals covering a wide range of parameters in noise, baseline and concept drifts. We show that the proposed approach enables to dissociate the discrete source signals emitted by the hidden molecular states from the continuous parasitic signal corresponding to the baseline drift, without any supervision nor prior knowledge on the sensor features or the underlying kinetics of the sensed phenomenon. We also demonstrate that our method is able to recover hidden molecular kinetic parameters, even under large noises and concept drifts. We report improved performances than model-based idealization in precision, recall, computational time, and than Bayesian non parametric approaches in terms of robustness to non-stationary and noisy signals. [1] C. Gu, C. Jia, and X. Guo, “Single-Molecule Electrical Detection with Real-Time Label-Free Capability and Ultrasensitivity,” Small Methods, vol. 1, no. 5, p. 1700071, 2017. [2] M. OUQAMRA and D. BOUILLY, “Unsupervised Drift Compensation Based on Information Theory for Single-Molecule Sensors,” , IEEE SigPort, 2019. [Online]. Available: http://sigport.org/4858. [3] Peter Grünwald, “Introducing the Minimum Description Length Principle,” in Advances in Minimum Description Length: Theory and Applications, chapter 1, pp. 3–22.MIT Press, 2005 [4] D. A. Huffman, “A Method for the Construction of Minimum-Redundancy Codes,” Proceedings of the IRE, vol. 40, pp. 1098–1101, Sept 1952.