· · · Repository hosted by TU Delft Library

A fully autonomous inductive boost converter for indoor photovoltaic harvesting with maximum power point tracking circuit is implemented in a commercial 0.25um CMOS process. The converter can handle input power from 5uW up to 10mW and charge a battery or a super-capacitor up to 5V. Its control circuit consumes between 0.8uA and 2.1uA depending on the input power level, resulting in a peak end-to-end efficiency of 70% when tracking a maximum input power of 17uW.

Purpose: X-ray based tomographic blood perfusion imaging requires recovery of contrast time-attenuation-curves from dynamic projection data. When using slowly rotating imaging systems this task is challenging due to non-simultaneous projection acquisition. A dynamic reconstruction method is proposed that aims at compensating the lack of simultaneously acquired information by incorporating prior knowledge about the expected temporal contrast dynamics. Methods: A decomposition model using temporal basis functions to approximate time-attenuation-curves is integrated into an iterative tomographicre construction method. The computationally efficient implementationof the proposed approach makes use of standard forward- and back projections as well as scalar products in image space. The critical issue of projection noise propagation is tackled by application of regularization which is realizedby early stopping of iteration cycles and by proper selection of smooth temporal basis functions. The performance of the proposed dynamic reconstruction approach is evaluated in a simulation study concerning various aspects: noise propagation and regularization, specification of temporal model, and type of acquisition mode. Results: The evaluation based on dynamic phantom data indicates that tomographic recovery of contrast time-attenuation-curves in tissue can be achieved with an average range of accuracy of ca. 2% (with respect todynamic peak attenuation) under ideal noise-free conditions. The relative estimation error for arterial time-attenuation-curves is in the range of 8%, which is due to faster contrast dynamics in the artery. In general, performance depends on the level of acquired information contained in the projection data which is mainly influenced by the type of rotational acquisition mode; restrictions in angular range and speed can lead to limited accuracy. The analysis of propagated projection noise in a statistical Bias-Variance framework reveals relative noise levels in estimated time-attenuation-curves of 3-4% intissue regions and below 1% in vessels when using optimized settingsfor regularization. Here, the effect of noise suppression depends oninterrelation between the model. Conclusions: For usage with slowly rotating imaging systems the presented model-based iterative dynamic reconstruction method is capable of recovering contrast time-attenuation-curves related to tissue perfusion. The proposed regularization framework is an effective means to limit the impact of projection noise which is a factor dominating estimation accuracy in tissue regions.

Many approaches to facial expression recognition focus on assessing the six basic emotions (anger, disgust, happiness, fear, sadness, and surprise). Real-life situations proved to produce many more subtle facial expressions. A reliable way of analyzing the facial behavior is the Facial Action Coding System (FACS) developed by Ekman and Friesen, which decomposes the face into 46 action units (AU) and is usually performed by a human observer. Each AU is related to the contraction of one or more specific facial muscles. In this study we present an approach towards automatic AU recognition enabling recognition of an extensive palette of facial expressions. As distinctive features we used motion flow estimators between every two consecutive frames, calculated in special regions of interest (ROI). Even though a lot has been published on the facial expression recognition theme, it is still difficult to draw a conclusion regarding the best methodology as there is no common basis for comparison. Therefore our main contributions reside in the comparison of different ROI selections proposed by us, different optical flow estimation methods, and also in the comparison of two spatial-temporal classification methods: Hidden Markov Models (HMM) and Dynamic Bayesian Networks (DBN). The classifiers have been trained and tested on the Cohn-Kanade database. The experiments showed that under the same conditions regarding initialization, labeling and sampling, both methods produced similar results, achieving the same recognition rate of 89% for the classification of facial AUs. Still, by enabling non-fixed sampling and using HTK, HMMs rendered a better performance of 93% suggesting that are better suited for the special task of AUs recognition.

Facial image analysis, including face detection, face recognition,facial expression analysis, facial demographic classification, and so on, is an important and interesting research topic in the computervision and image processing area, which has many important applications such as human-computer interaction and visual surveillance. Acritical step for successful facial image analysis is to derive an effective facial representation from the original face images. In recent years, Local Binary Patterns (LBP) has received increasing attention for facial description. This paper presents a comprehensive survey of LBP methodology in the context of facial image analysis. Different aspects are addressed, including recent development of LBP, LBP feature selection, its applications in different facial image analysis tasks and existing systems, and so on. With more than 100 papers reviewed, this is the first extensive review on LBP-based facial image analysis.

We investigate the potential to enhance breast cancer event predictors by exploiting subtype information. We do this with a two-stage approach that first determines a sample's subtype using a recent module-driven approach, and secondly constructs a subtype-specific predictor to predict a metastasis event within five years. Our methodology is validated on a large compendium of microarray breast cancer datasets,including 43 replicate array pairs for assessing subtyping stability. Note that stratifying by subtype strongly reduces the training set sizes available to construct the individual predictors, which may decrease performance. Besides sample size, other factors likeunequal class distributions and differences in the number of samplesper subtype, easily obscure a fair comparison between subtype-specific predictors constructed on different subtypes, but also between subtype specific and subtype a-specific predictors. Therefore, we constructed a completely balanced experimental design, in which none ofthe above factors play a role and show that subtype-specific eventpredictors clearly outperform predictors that do not take subtype information into account.

Accurate identification of physical activity types has been achieved in laboratory conditions using single-site accelerometers and classification algorithms. This methodology is then applied to free-living subjects to determine activity behaviour. This study aimed at analysing the reproducibility of the accuracy of laboratory-trained classification algorithms in free-living subjects during daily life. A support vector machine (SVM), a feed-forward neural network (NN) and a decision tree (DT) were trained with data collected by a waist-mounted accelerometer during a laboratory trial. The reproducibility of the classification performance was tested on data collected in daily life using a multiple-site accelerometer (IDEEA) augmented with an activity diary for 20 healthy subjects (age: 30 ± 9; BMI: 23.0 ± 2.6 kg/m2). Leave-one-subject-out cross-validation of the training data showed accuracies of 95.1 ± 4.3%, 91.4 ± 6.7% and 92.2 ± 6.6% for the SVM, NN and DT, respectively. All algorithms showed a significantly decreased accuracy in daily life as compared to the reference truth represented by the IDEEA and diary classifications (75.6 ± 10.4%, 74.8 ± 9.7%, and 72.2 ± 10.3%; p<0.05). In conclusion, cross-validation of training data overestimates the accuracy of the classification algorithms in daily life.

Source-gated transistors (SGTs) have potentially very high output impedance and low saturation voltages, which make them ideal as building blocks for high performance analog circuits fabricated in thin-film technologies. The quality of the saturation is greatly influenced by the design of the field-relief structure incorporated into the source electrode. Starting from measurements on self-aligned polysilicon structures, we show through numerical simulations how the field plate design can be improved. A simple source field plate around 1μm long situated several tens of nm above the semiconductor can increase the low-voltage intrinsic gain by more than two orders of magnitude and offers adequate tolerance to process variations in a moderately scaled thin-film SGT.

Stroke is the second cause of death and leading cause of disabilityworldwide. Less than 5% of ischemic stroke patients receive the state-of-the art treatment of a thrombolytic drug tPA, and only about 10% of these gain additional benefit from it. Ultrasound (US)-inducedmicrobubble (MB) cavitation has been shown to enhance the efficacy of the tPA drug or dissolve clots without tPA. Such a sonothrombolysis (STL) treatment requires monitoring and control of MB cavitation to ensure its reproducible efficacy and safety. This paper presents a prototype of a US cavitation imaging system. It is a part of an image-guided sonothrombolysis system based on a commercial US scanneriE33 (Philips Healthcare) with an imaging probe S5-1. Backscattereddata from insonified MBs is spectrally analyzed to identify the dominant cavitation state: ultraharmonics indicate Stable Cavitation (SC) and broadband noise indicates Inertial Cavitation (IC). Cavitation at lower levels (neither of SC or IC) are classified as Moderate Oscillations (MO). The system is evaluated in vitro and in vivo. Avessel phantom with Definity microbubbles was imaged through a water path and through a human temporal bone sample. In vivo experimentshave also been conducted for detecting cavitation in real time in the brain transcranially in two pigs. Cavitation images have also been obtained and processed offline on 17 pigs of a swine sonothrombolysis study. The lateral resolution of the system is approximately 3mm at a 6cm depth, and the axial resolution is 3cm for a 20µs pulselength. The maximum frame rate of the prototype system is 2Hz. Cavitation imaging allows assessing the relative importance of the different cavitation states (MO, SC and IC) in the treatment area inside the skull and their changes as a function of acoustic amplitude. Thetemporal evolution of cavitation can also be assessed, showing thatone 20us pulse destroys the majority of the MBs in the treatment area at MIs higher than 1. Such a therapy monitoring system will be critical for the reproducible safe and effective administration of STLtreatment for acute ischemic stroke.

Cloud computing is an emerging technology that is expected to support Internet scale critical applications which could be essential to the healthcare sector. Its scalability, resilience, adaptability, connectivity, cost reduction, and high performance features have high potential to lift the efficiency and quality of healthcare. However,it is also important to understand specific risks related to security and privacy that this technology brings. This paper focuses on a home healthcare system based on cloud computing. It introduces several use cases and draws an architecture based on the cloud. A comprehensive methodology is used to integrate security and privacy engineering process into the software development lifecycle. In particular,security and privacy challenges are identified in the proposed cloud-based home healthcare system. Moreover, a functional infrastructureplan is provided to demonstrate the integration between the proposed application architecture with the cloud infrastructure. Finally, the paper discusses several mitigation techniques putting the focus on patient-centric control and policy enforcement via cryptographic technologies, and consequently on digital rights management and attribute based encryption technologies.

Focusing of attention on a repetitive visual stimulation (RVS) at aconstant frequency, elicits the so called steady-state visual evokedpotential (SSVEP). This effect can be advantageously utilized in brain-computer interfaces (BCIs). SSVEP based BCIs can offer higher bitrates and require shorter training time as compared to other BCI modalities. Detection of the SSVEP from the EEG can be facilitated through spatial filtering (linear combination of the signals recorded at several electrodes). Literature offers several options to performthis. In this paper we propose a taxonomy to categorize these methods and we extensively evaluate them using 22 stimulation frequencies.We suggest improvements to existing methods to increase the SSVEP detection performance. We also consider practical aspects in the discussion of results.

We investigate the performance of detectors with M antennas (co-located or distributed) under Rayleigh fading, in terms of detection diversity. Rather than the high-SNR concept of diversity order common in the communications literature, we adopt the notion recently advocated by Daher and Adve in the radar community: the slope of the average probability of detection ( ¯ PD) vs. SNR curve at ¯ PD = 0.5. This definition is well suited to spectrum sensing, which invariably deals with low SNR levels. It is shown that the diversity order growsasM for an optimal centralized detector having access to all observations, whereas for the two distributed schemes considered (the multiantenna energy detector and the OR detector) it grows no faster √M.

In many practical applications we are interested in the extraction of only one desired signal out of a mixture of signals. A disadvantage of most blind extraction approaches proposed in the literature isthat they are inefficient in the sense that they also separate or extract undesired signals. To deal with this inefficiency we exploit an a priori guess of direction of arrival related parameters of the desired signal, which serves as a mold. Based on this mold we createlinear combinations of noise-free correlation matrices that are usedto construct a single matrix with a specific eigenstructure. The eigenvector that corresponds to the smallest eigenvalue of this matrixis the desired extraction filter. Finally it is shown that this approach paves the way to make the algorithm flexible in the utilization of additional a priori information.

While it has been widely proved that breastfeeding is the healthiestfeeding option for a baby and its mother, the mechanisms by which ababy removes milk from the breast are still not completely known. Partly this is due to the lack of tools to analyze images of the infant oral cavity during feeding automatically and quantitatively. In this paper we propose two methods for analyzing ultrasound videos toautomatically detect relevant events such sucking and swallowing ofmilk and to discriminate different types of tongue action during milk removal. The proposed algorithms provide, for the first time, quantitative indications of the type of activities carried out during breastfeeding by the baby, promising unprecedented advancements in thefield.

Previously proposed superdirective acoustic zooming techniques mainly focus on controlling the directivity-factor of the constructed beampattern as a function of the zoom-parameter. This means that thesezooming techniques are only consistent in the case of spherically isotropic interferences. In practical situations however, often directional interferences (mainly coming from a single direction) are present. To have a consistent behaviour of the acoustic zooming, we willpropose a new zooming technique that is based on a novel beampattern construction. The beampattern is constructed in such a way that for every angle, the response is monotonically increasing/decreasing in a consistent way with the zooming-parameter.

This letter presents a transmitter for sub-THz radiation (up to 160GHz), which consists of a nonlinear transmission line (NLTL) and anextremely wideband (EWB) slot antenna on a silicon substrate of lowresistivity (10 Ohms•cm). The fabrication was realized using a commercially available 65 nm CMOS process. On-wafer characterization of the whole transmitter, of the standalone EWB antenna and of the standalone NLTL is presented. Reflection measurements show that the standalone EWB antenna has a &#8722;10dB impedance bandwidth in the frequency bands 75-100 GHz and 220-325 GHz, which agrees very well with the simulation results. The simulated radiation patterns are also presented, indicating that the antenna has an ominidirectional radiation performance. The antenna shows also a maximum power gain of -9.5 dBi between 90 GHz and 120 GHz. The output power of the NLTL alone and of the NLTL integrated with the EWB antenna is measured up to 178GHz.

The diagnosis and treatment of malaria infection requires detectingthe presence of malaria parasite in the patient as well as identification of the parasite species. We present an image processing-basedapproach to detect parasites in microscope images of blood smear andan ontology-based classification of the stage of the parasite for identifying the species of infection. This approach is patterned after the diagnosis approach adopted by a pathologist for visual examination and hence is expected to deliver similar results. We formulateseveral rules based on the morphology of the basic components of a parasite namely chromatin dot(s) and cytoplasm to identify parasite stage and species. Numerical results are presented for data taken from various patients. Sensitivity of 88% and specificity of 95% is reported by evaluation of the scheme on 55 images.

Connected and interoperable healthcare system promises to reduce thecost of the healthcare delivery, increase its efficiency and enableconsumers to better engage with clinicians and manage their care. However at the same time it introduces new risks towards security andprivacy of personal health information, which are considered to beprominent impediments towards the realization of full benefits of connected health. In the connected healthcare system of future, different security technologies will be used to address security risks indifferent trust domains. For the domain of personal health and healthy lifestyle services, which is less trusted, additional security mechanisms (e.g. digital rights management) are required next to access control which is traditionally used in the professional medical domain. To realize the vision of connected health, all these mechanisms should be interoperable with each other. In this paper we providean interoperability framework which allows digital rights managementand access control systems to seamlessly work together through theuse of ontology.

Recent developments in the spectrum policy and regulatory domains allow more flexible and efficient use of spectrum, notably the releaseof National Broadband Plan, the publication of final rules for TV white spaces, and the ongoing proceeding for secondary use of the 2360–2400 MHz band for Medical Body Area Networks (MBANS). These important changes open up great opportunities for cognitive radio (CR) toenable and support a variety of emerging applications, ranging fromsmart grid, public safety, broadband cellular, to medical applications. This article presents a high-level view on how cognitive radio (primarily from a dynamic spectrum access perspective) would supportsuch applications, the benefits that cognitive radio would bring, and also some challenges that are yet to be resolved. We also highlight related standardization that uses cognitive radio technologies tosupport such emerging applications.

We propose in this paper an approach to quantifying the rate of atrophy of the brain of patients with Alzheimer’s disease. This approachis based on Computational Anatomy which al- lows the computation ofintermediate MR brain volumes be- tween the ones of regular scans.This increases dramatically the granularity of brain structure information, without requir- ing extra scans. We define two spaces: (i) the joint brain tissue deformation displacement vector magnitude andJacobian, (ii) the joint polar angles of the displacement vector. The shape of the distribution patterns in both spaces allow us to: (i)quan- tify atrophy rates of specific brain structures, such as, theven- tricles and the hippocampus, (ii) to build up models for the in- terpolation and extrapolation of atrophy rate parameters. The novelty of this approach is that it allows us to interpolate and extrapolate atrophy rate parameters computed from the two spaces, and thusderive precise models for patient diagnosis and/or prognosis. We tested this approach on a set of ADNI patients with diagnosed Alzheimer’s disease, mild cognitive impairment, and normal controls. This approach could also be used in the diagnosis of patients with other neurodegener- ative diseases, such as, frontal lobe dementia and Schizophre- nia.

Surveillance systems in shopping malls or supermarkets are usually designated for assuring safety and detecting abnormal behavior. We used the distributed video cameras system to design digital shopping assistants which assess the behavior of customers while shopping, detect when they need assistance, and offer their support in case thereis a selling opportunity. In this paper we propose a system for analyzing human behavior patterns related to products interaction, which could reveal the customer’s level of interest. We extracted discriminative features for basic action detection and analyzed differentstatistical and spatio-temporal classification methods, which capture relations between frames, features, and basic actions. Our experiments show that it is possible to accurately recognize different shopping related actions (85.7%) and discriminate between the proposed levels of interest in (88%) of the cases.