Inmaculada Mora-Jiménez

Time-Space Sampling and Mobile Device Calibration for WiFi Indoor Location Systems

Indoor location systems based on IEEE 802.11b (WiFi) mobile devices often rely on the received signal strength indicator to estimate the user position. Two key characteristics of these systems have not yet been fully analyzed, namely, the temporal and spatial sampling process required to adequately describe the distribution of the electromagnetic field in indoor scenarios; and the device calibration, necessary for supporting different mobile devices within the same system. By using a previously proposed nonparametric methodology for system comparison, we first analyzed the time-space sampling requirements for WiFi indoor location systems in terms of conventional sampling theory and system performance. We also proposed and benchmarked three new algorithms for device calibration, with increasing levels of complexity and performance. We conclude that feasible time and space sampling rates can be used, and that calibration algorithms make possible the handling of previously unknown mobile devices in the system.

Advanced support vector machines for 802.11 indoor location

Due to the proliferation of ubiquitous computing services, locating a device in indoor scenarios has received special attention during recent years. A variety of algorithms are based on Wi-Fi measurements of the received signal strength and estimate the relation between this one and position using previous measurements at known locations. This problem naturally fits in well with learning algorithms such as neural networks, or support vector machines (SVM). However, existing machine learning techniques do not significantly outperform other simpler techniques, such as k-nn. This is mainly due to the fact that these solutions do not include significant a priori information. In this paper, we propose a technique to enhance these algorithms by including certain a priori information within the learning machine, using the spectral information of the training set, and a complex output to take advantage of the cross information in the two dimensions of the location. Specifically, we modify a SVM algorithm to obtain three advanced methods incorporating this information: one using an autocorrelation kernel, another using a complex output, and a third one combining both. These algorithms are compared to the k-nn and an SVM with Gaussian kernel, showing that including the a priori information improves the location performance.

Nonparametric Model Comparison and Uncertainty Evaluation for Signal Strength Indoor Location

Indoor location (IL) using received signal strength (RSS) is receiving much attention, mainly due to its ease of use in deployed IEEE 802.11b (Wi-Fi) wireless networks. Fingerprinting is the most widely used technique. It consists of estimating position by comparison of a set of RSS measurements, made by the mobile device, with a database of RSS measurements whose locations are known. However, the most convenient data structure to be used and the actual performance of the proposed fingerprinting algorithms are still controversial. In addition, the statistical distribution of indoor RSS is not easy to characterize. Therefore, we propose here the use of nonparametric statistical procedures for diagnosis of the fingerprinting model, specifically: 1) A nonparametric statistical test, based on paired bootstrap resampling, for comparison of different fingerprinting models and 2) new accuracy measurements (the uncertainty area and its bias) which take into account the complex nature of the fingerprinting output. The bootstrap comparison test and the accuracy measurements are used for RSS-IL in our Wi-Fi network, showing relevant information relating to the different fingerprinting schemes that can be used.

Traffic sign segmentation and classification using statistical learning methods

Abstract Traffic signs are an essential part of any circulation system, and failure detection by the driver may significantly increase the accident risk. Currently, automatic traffic sign detection systems still have some performance limitations, specially for achromatic signs and variable lighting conditions. In this work, we propose an automatic traffic-sign detection method capable of detecting both chromatic and achromatic signs, while taking into account rotations, scale changes, shifts, partial deformations, and shadows. The proposed system is divided into three stages: (1) segmentation of chromatic and achromatic scene elements using L ⁎ a ⁎ b ⁎ and HSI spaces, where two machine learning techniques (k-Nearest Neighbors and Support Vector Machines) are benchmarked; (2) post-processing in order to discard non-interest regions, to connect fragmented signs, and to separate signs located at the same post; and (3) sign-shape classification by using Fourier Descriptors, which yield significant advantage in comparison to other contour-based methods, and subsequent shape recognition with machine learning techniques. Experiments with two databases of real-world images captured with different cameras yielded a sign detection rate of about 97% with a false alarm rate between 3% and 4%, depending on the database. Our method can be readily used for maintenance, inventory, or driver support system applications.

Heart Rate Turbulence Denoising Using Support Vector Machines

Heart rate turbulence (HRT) is a transient acceleration and subsequent deceleration of the heart rate after a premature ventricular complex (PVC), and it has been shown to be a strong risk stratification criterion in patients with cardiac disease. In order to reduce the noise level of the HRT signal, conventional measurements of HRT use a patient-averaged template of post-PVC tachogram (PPT), hence providing with long-term HRT indexes. We hypothesize that the reduction of the noise level at each isolated PPT, using signal processing techniques, will allow us to estimate short-term HRT indexes. Accordingly, its application could be extended to patients with reduced number of available PPT. In this paper, several HRT denoising procedures are proposed and tested, with special attention to support vector machine (SVM) estimation, as this is a robust algorithm that allows us to deal with few available time samples in the PPT. Pacing-stimulated HRT during electrophysiological study are used as a low-noise gold standard. Measurements in a 24-h Holter patient database reveal a significant reduction in the bias and the variance of HRT measurements. We conclude that SVM denoising yields short-term HRT measurements and improves the signal-to-noise level of long-term HRT measurements.

Heart Rate Variability on 7-Day Holter Monitoring Using a Bootstrap Rhythmometric Procedure

Heart rate variability (HRV) markers have been widely used to characterize the autonomous regulation state of the heart from 24-h Holter monitoring, but long-term evolution of HRV indexes is mostly unknown. A dataset of 7-day Holter recordings of 22 patients with congestive heart failure was studied. A rhythmometric procedure was designed to characterize the infradian, circadian, and ultradian components for each patient, as well as circadian and ultradian fluctuations. Furthermore, a bootstrap test yielded automatically the rhythmometric model for each patient. We analyzed the temporal evolution of relevant time-domain (AVNN, SDNN, and NN50), frequency-domain (LF, HF, HFn, and LF/HF), and nonlinear (α1 and SampEn) HRV indexes. Circadian components were the most significant for all HRV indexes, but the infradian ones were also strongly present in NN50, HFn, LF/HF, α1, and SampEn indexes. Among ultradian components that one corresponding to 12 h, was the most relevant. Long-term monitoring of HRV conveys new potentially relevant rhythmometric information, which can be analyzed by using the proposed automatic procedure.

Predicting colorectal surgical complications using heterogeneous clinical data and kernel methods

OBJECTIVE In this work, we have developed a learning system capable of exploiting information conveyed by longitudinal Electronic Health Records (EHRs) for the prediction of a common postoperative complication, Anastomosis Leakage (AL), in a data-driven way and by fusing temporal population data from different and heterogeneous sources in the EHRs. MATERIAL AND METHODS We used linear and non-linear kernel methods individually for each data source, and leveraging the powerful multiple kernels for their effective combination. To validate the system, we used data from the EHR of the gastrointestinal department at a university hospital. RESULTS We first investigated the early prediction performance from each data source separately, by computing Area Under the Curve values for processed free text (0.83), blood tests (0.74), and vital signs (0.65), respectively. When exploiting the heterogeneous data sources combined using the composite kernel framework, the prediction capabilities increased considerably (0.92). Finally, posterior probabilities were evaluated for risk assessment of patients as an aid for clinicians to raise alertness at an early stage, in order to act promptly for avoiding AL complications. DISCUSSION Machine-learning statistical model from EHR data can be useful to predict surgical complications. The combination of EHR extracted free text, blood samples values, and patient vital signs, improves the model performance. These results can be used as a framework for preoperative clinical decision support.

Symmetrical Compression Distance for Arrhythmia Discrimination in Cloud-Based Big-Data Services

The current development of cloud computing is completely changing the paradigm of data knowledge extraction in huge databases. An example of this technology in the cardiac arrhythmia field is the SCOOP platform, a national-level scientific cloud-based big data service for implantable cardioverter defibrillators. In this scenario, we here propose a new methodology for automatic classification of intracardiac electrograms (EGMs) in a cloud computing system, designed for minimal signal preprocessing. A new compression-based similarity measure (CSM) is created for low computational burden, so-called weighted fast compression distance, which provides better performance when compared with other CSMs in the literature. Using simple machine learning techniques, a set of 6848 EGMs extracted from SCOOP platform were classified into seven cardiac arrhythmia classes and one noise class, reaching near to 90% accuracy when previous patient arrhythmia information was available and 63% otherwise, hence overcoming in all cases the classification provided by the majority class. Results show that this methodology can be used as a high-quality service of cloud computing, providing support to physicians for improving the knowledge on patient diagnosis.

Ontology for Heart Rate Turbulence Domain From The Conceptual Model of SNOMED-CT

Electronic health record (EHR) automates the clinician workflow, allowing evidence-based decision support and quality management. We aimed to start a framework for domain standardization of cardiovascular risk stratification into the EHR, including risk indices whose calculation involves ECG signal processing. We propose the use of biomedical ontologies completely based on the conceptual model of SNOMED-CT, which allows us to implement our domain in the EHR. In this setting, the present study focused on the heart rate turbulence (HRT) domain, according to its concise guidelines and clear procedures for parameter calculations. We used 289 concepts from SNOMED-CT, and generated 19 local extensions (new concepts) for the HRT specific concepts not present in the current version of SNOMED-CT. New concepts included averaged and individual ventricular premature complex tachograms, initial sinus acceleration for turbulence onset, or sinusal oscillation for turbulence slope. Two representative use studies were implemented: first, a prototype was inserted in the hospital information system for supporting HRT recordings and their simple follow up by medical societies; second, an advanced support for a prospective scientific research, involving standard and emergent signal processing algorithms in the HRT indices, was generated and then tested in an example database of 27 Holter patients. Concepts of the proposed HRT ontology are publicly available through a terminology server, hence their use in any information system will be straightforward due to the interoperability provided by SNOMED-CT.

Implantable Defibrillator Electrograms and Origin of Left Ventricular Impulses: An Analysis of Regionalization Ability and Visual Spatial Resolution

ICD Electrograms and Origin of Impulses. Introduction: The implantable cardioverter‐defibrillator (ICD) electrogram (EG) is a documentation of ventricular tachycardia. We prospectively analyzed EGs from ICD electrodes located at the right ventricle apex to establish (1) ability to regionalize origin of left ventricle (LV) impulses, and (2) spatial resolution to distinguish between paced sites. Methods and Results: LV electro‐anatomic maps were generated in 15 patients. ICD‐EGs were recorded during pacing from 22 ± 10 LV sites. Voltage of far‐field EG deflections (initial, peak, final) and time intervals between far‐field and bipolar EGs were measured. Blinded visual analysis was used for spatial resolution. Initial deflections were more negative and initial/peak ratios were larger for lateral versus septal and superior versus inferior sites. Time intervals were shorter for apical versus basal and septal versus lateral sites. Best predictive cutoff values were voltage of initial deflection <–1.24 mV, and initial/peak ratio >0.45 for a lateral site, voltage of final deflection <–0.30 for an inferior site, and time interval <80 milliseconds for an apical site. In a subsequent group of 9 patients, these values predicted correctly paced site location in 54–75% and tachycardia exit site in 60–100%. Recognition of paced sites as different by EG inspection was 91% accurate. Sensitivity increased with distance (0.96 if ≥ 2 cm vs 0.84 if < 2 cm, P  <  0.001) and with presence of low‐voltage tissue between sites (0.94 vs 0.88, P  <  0.001). Conclusions: Standard ICD‐EG analysis can help regionalize LV sites of impulse formation. It can accurately distinguish between 2 sites of impulse formation if they are ≥2 cm apart. (J Cardiovasc Electrophysiol, Vol. 23, pp. 506‐514, May 2012)