No abstract

The ability to accurately stratify patients at risk of adverse cardiovascular outcomes using heart sound recordings could result in earlier treatment and improved patient outcomes. However, there remain several challenges associated with risk stratifying patients based on the phonocardiogram (PCG) alone. First, inter-patient differences can make it challenging to learn a model that generalizes well across patients. Second, heterogeneity introduced by the collection environment of the recordings can render a classifier trained on one population useless when applied to another. To address these challenges we explore the use of temporal alignment techniques, in particular dynamic time warping (DTW). Using DTW we compare heart sounds within and across subjects/recordings. These DTW based features, coupled with widely used spectral MFCC coefficients, serve as input to a linear SVM. Applied to the held-out test set our classifier obtained a test score of 82.4%, suggesting that temporal alignment techniques can effectively reduce the effects of inter-patient variability and mitigate the differences introduced by heterogeneous data collection environments. IntroductionIn cardiac auscultation an examiner uses a stethoscope to listen for unique and distinct sounds, that provide important data regarding the condition of the heart. Modern recording equipment captures these heart sounds as a phonocardiogram (PCG). In principle, these recordings could be used to automatically monitor patients and diagnose cardiac abnormalities. Yet, while auscultation is a common practice in patient exams, PCGs are not widely used clinically, where echocardiograms and electrocardiograms are more prevalent. This is due, in part, to the lack of robust algorithms for automatically classifying PCGs. To address this issue, the 2016 PhysioNet/CinC Challenge focused on the development of algorithms to classify PCGs collected from both clinical and nonclinical environments [1].Robust PCG classification algorithms must accurately identify cardiac abnormalities across patients and across diverse recording environments. To address challenges associated with inter-patient variability we borrow techniques that have been successfully applied in speech processing and ECG analysis, where similar issues arise [2][3][4]. In particular, we explore the use of dynamic time warping (DTW) in measuring similarity between heartbeats from the same subject and across subjects. Our experiments show that such DTW-based features can mitigate the differences introduced by heterogeneous data collection environments and improve classification performance, especially when training and test populations differ. MethodsIn this section we present our supervised learning system for classifying PCGs as either normal of abnormal. We begin by describing the signal segmentation, then move on to feature extraction and lastly explain the learning algorithm. SegmentationAs a first step, we segment the PCG recording into the fundamental heart sounds: S1 and S2 in addition to the s...

This paper presents a novel framework for objects detection in security and broadcast videos. Our method assumes thatobject classes are unknown in advance and exploit the temporal-space properties of the videos for the creation of avocabulary that describes these classes. Local space-time features have recently became a popular video representationfor action recognition and object detection. Several methods for feature localization and description have been proposedin the literature and promising recognition results were demonstrated for a number of action classes.In this work we propose the use of different kinds of descriptors for the creation of vocabularies for different detectionobject task. For a better description of the videos we carry out a background model, tryring to clean up and follow theareas where there are objects. The points of interest in the videos to characterize the objects are calculated with atemporary variant of the famous Harris corner detector. With the descriptors obtained from the points of interest, avocabulary is realized usingthe kinds of videos we want to train. Then we obtained the frequency histogramsbetween the videos for training and the vocabulary so, with a binary classifier obtain the trained classes and followingthe same procedure without the vocabulary realized the detection and monitoring of the objects.The new method presented is also compared with a state of the art method, obtaining better results in both accuracyand false object rejection.

A space-filling curve in 2, 3, or higher dimensions can be thought as a path of a continuously moving point. As its main goal is to preserve spatial proximity, this type of curves has been widely used in the design and implementation of spatial data structures and nearest neighbor-finding techniques. This paper is essentially focused on the efficient representation of Digital Elevation Models (DEM) that entirely fit into the main memory. We propose a new hierarchical quadtree-like data structure to be built over domains of unrestricted size, and a representation of a quadtree and a binary triangles tree by means of the Hilbert and the Sierpinski space-filling curves, respectively, taking into account the hierarchical nature and the clustering properties of this kind of curves. Some triangulation schemes are described for the space-filling-curves-based approaches to efficiently visualize multiresolution surfaces.