Distinguished Lecturer Program

In this talk, we discuss how computer vision can facilitate the interpretation of medical imaging data,  or help  making  inferences  based  on models  of such  data.  In order  to illustrate this presentation, several applications of medical imaging measurements and modeling are discussed, focusing in areas such as the correction of imaging artifacts that may occlude visual information, tumor detection, modeling and measurement in different imaging modalities.

When interpreting medical imaging data with computer vision, usually we are trying to describe anatomic structures (or medical phenomena) using one or more images, and reconstruct some of its properties based on imaging data (like shape, texture or color). Actually, this is an ill-posed problem that humans can learn to solve effortlessly, but computer algorithms often are prone to errors. Nevertheless, in some cases computers can surpass humans and interpret medical images more accurately, given the proper choice of models, as we will show in this talk.

Reconstructing interesting properties of real world objects or phenomena from captured imaging data involves solving an inverse problem, in which we seek to recover some unknowns given insufficient information to specify a unique solution. Therefore, we disambiguate   between   possible   solutions   relying   on   models   based   on   physics, mathematics or statistics. Modeling the real world in all its complexity still is an open problem. However, if we know the phenomenon or object of interest, we can construct detailed models using specialized techniques and domain specific representations, that are efficient at describing reliably the measurements (or obtaining measurements in some cases). In this talk, we briefly overview some challenging problems in computer vision for medical imaging and measurements, with illustrations and insights about model selection and model-based prediction. Some of the applications discussed in this talk are: modeling  tumor  shape and  size,  and  making  inferences  about  its  future  growth  or shrinkage; modeling relevant details in the background of medical images to discriminate them from useless background noise; and modeling shading artifacts to minimize their influence when detecting and measuring skin lesions in standard camera images.

Medical  images  contain  a wealth  of information,  which  makes  modeling of medical images a challenging task. Therefore, medical images often are segmented into multiple elementary parts, simplifying their representation and changing the image model into something that is more meaningful, or easier to analyze and measure (e.g. by describing the objects boundaries by lines or curves, or the image segments by their textures, colors, etc.). Nevertheless, these simpler image elements may be easy to perceive visually but difficult  to  describe.  For  example,  the  texture  of  a skin lesion may  not  have  an identifiable texture element or a model known a priori, and regardless of that skin lesion detection   must  be  accurate   and  precise.   Segmentation   of  medical imaging  data segmentation and analysis still is an open question, and some current directions are discussed in this talk.

Computer vision and modeling are interrelated. Modeling imaging measurements often involves errors, and estimating the expected error of a model can be important in applications  (e.g. estimating a  tumor  size  and  its  potential  growth,  or  shrinkage,  in response to a treatment). This issue can be approached by adapting machine learning and pattern recognition techniques to solve problems in medical imaging measurements. Typically, a model has tuning parameters, and these tuning parameters may change the model complexity. We wish to minimize modeling errors and the model complexity, in other words, to get the ‘big picture’ we often sacrifice some of the small details. For example, estimating tumor growth (or shrinkage) in response to treatment requires modeling the tumor shape and size, which can be challenging for real tumors, and simplified models may be justifiable if the predictions obtained are informative (e.g. to evaluate the treatment effectiveness). To conclude this talk, we outline the current trends in computer vision in medical imaging measurements, and discuss some open problems.

Abstract - Microwave and millimeter-wave signals span the frequency range of ~300 MHz to 300 GHz, corresponding to a wavelength range of 1000 mm to 1 mm. Signals at these frequencies can easily penetrate inside dielectric materials and composites and interact with their inner structures. The relatively small wavelengths and wide bandwidths associated with these signals enable the production of high spatial-resolution images of materials and structures. Incorporating imaging techniques such as lens-focused and near-field techniques, synthetic aperture focusing, holographical methods based on robust back-propagation algorithms with more advanced and unique millimeter wave imaging systems have brought upon a flurry of activities in this area and in particular for nondestructive evaluation (NDE) applications.  These imaging systems and techniques have been successfully applied for a wide range of critical NDE-related applications.

Although, near-field techniques have also been prominently used for these applications in the past, undesired issues related to changing standoff distance and slowness of image production process have resulted in several innovative and automatic standoff distance variation removal techniques. Ultimately, imaging techniques must produce high-resolution 3D images, become real-time, and be implemented using portable systems.  To this end and to expedite the imaging process while providing a high-resolution images, the design and demonstration of a 6” by 6” one-shot, rapid and portable imaging system (Microwave Camera), consisting of 576 resonant slot elements, was demonstrated a few years ago.  Subsequently, efforts were expended to design and implement several different variations of this imaging system to accommodate one-sided and mono-static imaging, while enabling 3D image production using non-uniform rapid scanning of an object, as well as increasing the operating frequency into higher millimeter wave frequencies. These efforts have led to the development of a real-time, portable, high-resolution and 3D imaging microwave camera operating in the 20-30 GHz frequency range which was recently completed. This presentation provides an overview of these techniques, along with illustration of several typical examples where these imaging techniques have effectively provided viable solutions to many critical NDE problems.  

Semiconductor chip manufacturing cost consists of die cost, package cost, and test cost. The trends of increasing design complexity, increasing quality needs, and new process nodes and defect models are pushing test cost to the forefront. This is especially true for high-resolution data convertors, whose accurate testing requires expensive instruments and is extremely time-consuming. As a result, linearity test of data convertors often dominates the overall test cost of SoCs. This talk will present several recently developed techniques for reducing linearity test cost by dramatically reducing measurement time and dramatically relaxing instrumentation requirements.

The IEEE standard for ADC linearity test requires the stimulus signal to be at least 10 times more accurate than the ADC under test. To relax this stringent requirement, the SEIR (stimulus error identification and removal) algorithm is developed to accurately test high resolution ADCs using nonlinear stimuli. It has been demonstrated by industries that more than 16 bits of ADC test accuracy were achieved using 7-bit linear ramps instead of 20-bit linear ramps as required by IEEE, a relaxation of well over 1000 times on the instrumentation accuracy requirement.

The biggest contributor to test cost is the long measurement time. The recently developed uSMILE (ultrafast Segmented Model Identification for Linearity Errors) algorithm can dramatically reduce the measurement time needed for ADC linearity test. With a system identification approach using a segmented model for the integral nonlinearity, the algorithm can reduce the test time by a factor of over 100 and still achieve test accuracies superior to the standard histogram test method. This method has been extensively validated by industry and has been adopted for production test for multiple product families.

By combining the salient features of both SEIR and uSMILE, the ultrafast stimulus error removal and segmented model identification of linearity errors (USER- SMILE) algorithm is developed. The USER-SMILE algorithm uses two nonlinear signals as input to the ADC under test. One signal is shifted by a constant voltage with respect to the other nonlinear signal. By subtracting the two sets of output codes, input signal is canceled and the nonlinearity of ADC, modeled by a segmented non-parametric INL model, will be identified with the least square method.

A completely on-chip ADC BIST circuit is developed based on the USER-SMILE algorithm and demonstrated on a 28nm CMOS automotive microcontroller. The ADC test subsystem includes a nonlinear DAC as signal generator, a built-in voltage shift generator, a BIST computation engine, and dedicated memory cells. The silicon measurement results show accurate test results. The INL test results are further used to correct ADC linearity errors, thus providing a method for reliably calibrating the ADC. Measurement results demonstrated that the BIST-based calibration method achieved >10dB THD/SFDR improvements over the existing calibration method used by industry.

The tutorial will focus on sensor and measurement systems for new generations of vehicles with driver-assisted/autonomous capability. This is the main trend that is revolutionizing vehicles and mobility of people and goods, and is also making smart our cities. The economic and social impacts of this application field are huge. Worldwide every year 90 millions of vehicles are sold, but 1.25 millions of people are killed due to lack of safety. In US 3.1 billions of gallons of fuel are wasted due to traffic congestion. Assisted driving and autonomous driving aim at increasing safety, at improving fuel efficiency and our lifestyle by avoiding traffic congestion, at ensuring mobility for elderly and disabled people (inclusivity). The interest in this research subject is demonstrated by the huge investments of companies like Google, Intel, Tesla, Uber, Ford, GM, to name just a few, and by technology alliances, e.g. between BMW and Intel, planning autonomous cars for 2021. A convergence between automotive and ICT/Electronics industry is foreseen in the near future. An example of this convergence is the 5G Automotive Association http://www.5gaa.org/, which includes all main cars’ manufacturers, telecom service providers, electronic industries, measurement system providers (Keysight, Rohde&Schwarz).

The key enabling technologies for this scenario are the sensing and measurement systems, needed for the accurate vehicle positioning and navigation, for vehicle context-awareness, obstacle detection and collision avoidance, for driver-assistance (enhanced vision, driver’s attention and fatigue detection).

The lecture will be divided in multiple sections.

First, in the Introduction, innovation and market trends in the field of sensor and measurement technologies applied to vehicles and smart mobility systems will be discussed, focusing on next generation of driver-assisted/autonomous vehicles.

Then, new Radar and Lidar systems, appearing on-board vehicles beside array of imaging cameras, will be discussed for measurement of obstacle positions, distance and relative speed. A trade-off has to be found between power and size of active sensing systems like Radar and Lidar and their maximum measurement range. Moreover, in continuous wave Radars the limited frequency sweep range and the limited number of TX/RX channels lead to limits for the resolution in distance, direction of arrival, and speed measurements. Examples of X-band mobility surveillance Radar and mm-wave automotive Radar will be provided.

On the other hand, MOEMS (micro opto electro mechanical systems)-based scanned systems, used to reduce size and cost of Lidars are causing distortions that are worsening the accuracy of light-based measurements. Distortions due to fish-eye lenses, used to enlarge the field-of-view, are decreasing measurement performance of imaging sensors. Techniques to mitigate such artefacts will be discussed.

Practical examples of traffic sign recognition systems, road signs recognition, image mosaicking for all around view will be discussed. In addition, Lidar and imaging cameras suffer of decreased measurement performance in case of harsh operating conditions (e.g. bad weather or light conditions).

New biometric sensing and measurement systems will be also reviewed, such as Radar-based contactless heart/breath-rate measurement, smart steering-wheel for skin temperature/galvanic-response measurements or heart-rate detection, with the final aim of detecting the driver’s attention or health status.

Concerning on-board sensors for positioning and navigation, recent advances in MEMS accelerometers and gyroscope will be discussed. A careful analysis will be carried out about the measurement errors they cause on position and navigation, due to their bias and random walk output noise.

Finally, the lecture will analyze the trend in computing platforms, where parallel architectures and machine learning/AI (artificial intelligence) techniques, will be exploited to manage in real-time many and heterogeneous sources of measurements and to take autonomous decisions.

Suggestions for future directions of interest for the I&M society, and references to recent publications on IMS journals and conferences, in the field of automated and connected vehicles, will be provided as a conclusion.

Scientific and industrial worlds have started recently to look again with interest to the basic rules to perform reliability, availability and safety analysis and design on complex electro-mechanical systems. The main failure modes on electronic devices and sensors as well as the main techniques for failure mode investigation are of interest in modern system design. Statistical characterization of the main probability density functions and degradation models of innovation is mandatory to build lasting and safe products. The main reliability design techniques such as: fault tree analysis, cut set method, minimal path approach, critical block analysis for reliability are requested by companies worldwide as well as the knowledge of the main failure modes and reliability databases and handbooks as MIL-HDBK217, OREDA, BELLCORE, etc… Maintenance policies with special attention to corrective and preventive ones are also affected by reliability design in terms of advantages and disadvantages when applied to electro-mechanical systems. The main safety standards as IEC61508, IEC 61511 and EN50129, EN50128, EN50126 are usually considered in industrial design. The aim of this talk is to enable companies to develop inner confidence on advanced modelling techniques involving reliability, availability and safe design. Under this spotlight in addition to traditional and well known statistical models, innovative modelling techniques based on statistical data representation will be introduced and tailored to some specific case studies in the fields of bio instruments, transportations and oil & gas contexts.     

Abstract: Modern control algorithms in the emerging power systems process information delivered mainly by distributed, synchronized measurement systems, and available in data streams with different reporting rates. Multiple measurement approaches are used: on one side, the existing time-aggregation of measurements are offered by currently deployed IEDs (SCADA framework), including smart meters and other emerging units; on the other side, the high-resolution waveform-based monitoring devices like phasor measurement units (PMUs) use high reporting rates (50 frames per second or higher) and can include fault-recorder functionality.

There are several applications where synchronized data received with high reporting rate has to be used together with aggregated data from measurement equipment having a lower reporting rate (complying with power quality data aggregation standards) and the accompanying question is how adequate are the energy transfer models in such cases. For example, state estimators need both types of measurements: the so-called “classical” one, adapted for a de facto steady-state paradigm of relevant quantities and the “modern” one, i.e. with fewer embedded assumptions on the variability of same quantities. Another example is given by emerging active distribution grids operation, which assumes higher variability of the energy transfer and consequently a new model approximation for its characteristic quantities (voltages, currents) is needed. Such a model is required not only in order to be able to correctly design future measurement systems but also for better assessing the quality of existing “classical” measurements, still in use for power quality improvement, voltage control, frequency control, network parameters’ estimation etc.

The main constraint so far is put by the existing standards where several aggregation algorithms are recommended, with specific focus on the information compression. The further processing of rms values (already the output of a filtering algorithm) results in significant signal distortion.

Presently there is a gap between (i) the level of approximation used for modeling the current and voltage waveforms which is implicitly assumed by most of the measurement devices deployed in power systems and (ii) the capabilities and functionalities exhibited by the high fidelity, high accuracy and high number of potential reporting rates of the newly deployed synchronized measurement units.

The talk will address:

o The measurement paradigm in power systems;

• System inertia, real time and steady-state
• Instrument transformers; limited knowledge on the infrastructure
• PQ, SCADA and PMUs
• Power system state estimation; WAMCS
• IEDs, PMUs, microPMUs
• Time-stamped versus synchronized measurements

o Measurement channel quality and models for energy transfer

• Voltage and frequency variability; rate of change of frequency
• The steady-state signal and rapid voltage changes (RVC); rms-values reported with 100 frames/s;
• Measurement data aggregation; filtering properties
• Time- aggregation algorithms in the PQ framework
• Statistical approaches;

o Applications and challenges

• Communication channel requirements; delay assessment in WAMCS
• Smart metering with high reporting rate (1s)

The presentation provides an overview of these techniques, with examples from worldwide measurement solutions for smart grids deployment.