Analysis of photosensor frontends for biomedical applications: a focus on noise

Photosensing has proven to be highly useful in biomedical research and medical care. Being able to detect and quantify the light emitted by fluorescent or bioluminescent reporters has significantly participated in the development of new cancer treatments and made possible the development of cost-effective diagnosis techniques. While these advances relied on the large CCD devices or on laboratory-on-chips, the design of a bioluminescence-based implant for the continuous monitoring of biological data can be considered and would allow for the acquisition of insightful information in both research and care. To exist, such an implant requires a suited photosensing circuit, the design of which can be challenging due to the weakness of the target bioluminescent signal. This thesis addresses the design and evaluation of an appropriate analog-to-digital converter (ADC), with a focus on its expected accuracy. In particular, three integration-base current-to-frequency converters are considered: the current-controlled oscillator, the first-order Sigma-Delta converter with instantaneous feedback, the first-order Sigma-Delta converter with continuous feedback. Theoretical formulas are established to assess how the input-referred noise would impact the results from such ADCs, depending on the integration time as well as the time passed since last calibration. In addition to the general case, specific formulas are provided for white noise, flicker noise, and Lorentzian-shaped noise. Moreover, in practical cases that are conceivable with target application, these formulas cannot be used as they are due to the finite resolution of computing softwares. Consequently, the related implementation issues are addressed and solved. To guide the design of the current-to-frequency converters, how each of their imperfections affects their accuracy is analyzed in depth. This analysis and the obtained formulas serve a top-level-focused design strategy that is argued for in this thesis. It relies on maintaining a detailed top-level description of the system under study to help focus the designing efforts on the most impactful areas as well as guide the design choices at the transistor level. Applying such a strategy is made possible by the development of a MATLAB tool providing a flexible graphic user interface to explore the impact of each parameter of the design. In addition, to further assist the design of the ADC and/or of the photodetector, a Python tool is presented to explore configurations for successive simulations and data analyses, even when using multiple softwares. As an opening chapter, emphasizing the importance of considering the top-level in design, the case of heart rate monitoring through photoplethysmography is addressed. In this case, the highest level of the system includes the data processing, and only the latter can actually allow for drastic improvements of the power consumption by alleviating the requirements on the data acquisition. In fact, it is demonstrated that in many situations, the sampling frequency for the acquisition of the plethysmograph could be brought down to 10 to 25 Hz, while still ensuring a satisfying accuracy for the heart rate and heart rate variation monitoring.

Adaptive wake and sleep detection for wearable systems

Walter Karlen

Sleep problems and disorders have a serious impact on human health and wellbeing. The rising costs for treating sleep-related chronic diseases in industrialized countries demands efficient prevention. Low-cost, wearable sleep / wake detection systems which give feedback on the wearer's "sleep performance" are a promising approach to reduce the risk of developing serious sleep disorders and fatigue. Not all bio-medical signals that are useful for sleep / wake discrimination can be easily recorded with wearable systems. Sensors often need to be placed in an obtrusive location on the body or cannot be efficiently embedded into a wearable frame. Furthermore, wearable systems have limited computational and energetic resources, which restrict the choice of sensors and algorithms for online processing and classification. Since wearable systems are used outside the laboratory, the recorded signals tend to be corrupted with additional noise that influences the precision of classification algorithms. In this thesis we present the research on a wearable sleep / wake classifier system that relies on cardiorespiratory (ECG and respiratory effort) and activity recordings and that works autonomously with minimal user interaction. This research included the selection of optimal signals and sensors, the development of a custom-tailored hardware demonstrator with embedded classification algorithms, and the realization of experiments in real-world environments for the customization and validation of the system. The processing and classification of the signals were based on Fourier transformations and artificial neural networks that are efficiently implementable into digital signal controllers. Literature analysis and empiric measurements revealed that cardiorespiratory signals are more promising for a wearable sleep / wake classification than clinically used signals such as brain potentials. The experiments conducted during this thesis showed that inter-subject differences within the recorded physiological signals make it difficult to design a sleep / wake classification model that can generalize to a group of subjects. This problem was addressed in two ways: First by adding features from another signal to the classifier, that is, measuring the behavioral quiescence during sleep using accelerometers. Conducted research on different feature extraction methods from accelerometer data showed that this data generalizes well for distinct subjects in the study group. In addition, research on user-adaptation methods was conducted. Behavioral sleep and wake measures, notably the measurement of reactivity and activity, were developed to build up a priori knowledge that was used to adapt the classification algorithm automatically to new situations. This thesis demonstrates the design and development of a low-cost, wearable hardware and embedded software for on-line sleep / wake discrimination. The proposed automatic user-adaptive classifier is advantageous compared to previously suggested classification methods that generalize over multiple subjects, because it can take changes in the wearer's physiology and sleep / wake behavior into account without adjustment from a human expert. The results of this thesis contribute to the development of smart, wearable, bio-physiological monitoring systems which require a high degree of autonomy and have only low computational resources available. We believe that the proposed sleep / wake classification system is a first promising step toward a context-aware system for sleep management, sleep disorder prevention, and reduction of fatigue.

EPFL2009

Active: a unified platform for building intelligent applications

Didier Guzzoni

Computers have become affordable, small, omnipresent and are often connected to the Internet. However, despite the availability of such rich environments, user interfaces have not been adapted to fully leverage their potential. In our view, user interfaces will evolve to become more than simple tools that act using a click-and-act paradigm – people want software that can act as assistants to whom tasks can be delegated. This new type of software will provide us with more user-centric systems, able to interact naturally with human users and with the information environment. Although progress has been made in this direction over the last decade, current research in the field has shown that building intelligent assistants is a complex task that requires expertise in many fields, ranging from artificial intelligence to core software and hardware engineering. The difficulty of deeply integrating all the technologies and AI methodologies required to produce robust intelligent assistants has greatly limited the impact and widespread adoption of this form of software. In this thesis, we propose to design, implement and evaluate a new methodology and associated tool suite aimed to ease and accelerate the development of intelligent assistant software. Active, our solution, introduces the original concept of Active Ontologies, and combines it with a service oriented architecture to serve as the foundation for user-centric applications. The Active software suite features a programming editor, a runtime server and associated management tools. Using this unified platform, we developed techniques for rapidly creating intelligent assistant applications that weave together language processing, process modeling and dynamic service orchestration. To validate our approach, three prototypes have been implemented and evaluated. First, we created an assistant that helps mobile users retrieve data and access online services. The system allows users to fetch information through natural language dialog about restaurants, hotels, points of interest, flights status and weather forecasts. As a second prototype, we have implemented an assistant to help surgeons in the context of the operating room. The system uses a combination of voice and gesture recognition to help surgeons navigate through pre-operative information, visualize and control a live video stream coming from an endoscope mounted on a robotic arm. Lastly, we have created a system that helps organize meetings through emails and instant messages with the organizer and attendees. These three prototypes are deployed in very different application domains, and yet are built with the same tools and methods; this demonstrates the flexibility and versatility of our approach. We conducted a user study to validate our claim that, in specific domains, intelligent assistant software can perform better than conventional software approaches. First, we compared our mobile assistant against Google Mobile™, the leading commercially available search application for mobile users. We asked a population of users to accomplish ten travel-related tasks with both systems. Results show that, for the travel domain, our assistant-based system performs significantly better both in terms of effectiveness and time to completion than the more conventional, keyword-based search engine. In the medical domain, we asked a small population of surgeons to accomplish a set of tasks with our second prototype. After providing us with a positive feedback, surgeons used our prototype to discuss and experiment with innovative forms of user interaction in the context of the operating room. We also performed evaluations to support our claim that our unified approach accelerates the software development of intelligent assistants. First, we gave basic training on the Active platform to a population of software developers, asking them to go through a simple tutorial. Once trained, they were able to successfully program the core component of the mobile assistant prototype in under two hours. As a second validation, the meeting organizing assistant was designed, implemented and tested within three working days – comparable systems published in the literature were the product of significantly more time and resource investment. This demonstrates that a system encompassing numerous AI and software development aspects, including language processing, multiple modalities, contextual dialog, processing logic and backend services integration, can be rapidly implemented and tested using the Active framework. Our work shows that for many task-oriented domains, intelligent assistant applications can be significantly more effective than conventional systems. Through our Active Ontology approach and associated methodologies and tools, we have shown that complex intelligent assistant applications, perhaps for the first time, can be developed with low cost and high reliability. In addition to receiving a positive echo from the research world, our work is also causing ripples in the commercial world.

EPFL2008

Structure-preserving approaches and data-driven closure modeling for model order reduction

Nicolò Ripamonti

In this thesis, we propose model order reduction techniques for high-dimensional PDEs that preserve structures of the original problems and develop a closure modeling framework leveraging the Mori-Zwanzig formalism and recurrent neural networks. Since high-fidelity approximations of PDEs often result in a large number of degrees of freedom, the need for iterative evaluations for numerical optimizations and rapid feedback is computationally challenging.The first part of this thesis is devoted to conserving the high-dimensional equation's invariants, symmetries, and structures during the reduction process. Traditional reduction techniques are not guaranteed to yield stable reduced systems, even if the target problem is stable. In the context of fluid flows, the skew-symmetric structure of the problem entails the preservation of the kinetic energy of the system. By preserving the same structure at the level of the reduced model, we obtain enhanced stability, and accuracy and the reduced model acquires physical significance by preserving a surrogate of the energy of the original problem. Next, we focus on Hamiltonian systems, which, being driven by symmetry, are a source of great interest in the reduction community. It is well known that the breaking of these symmetries in the reduced model is accompanied by a blowup of the system energy and flow volume. In this thesis, geometric reduced models for Hamiltonian systems are further developed and combined with the dynamically orthogonal methods, addressing the poor reducibility in time of advection-dominated problems. The reduced solution is expressed as a linear combination of a finite number of modes and coincides with the symplectic projection of the high-fidelity Hamiltonian problem onto the tangent space of the approximating manifold. An error surrogate is used to monitor the approximation ability of the reduced model and make a change in the rank of the approximating system if necessary. The method is further developed through a combination of DEIM and DMD to reduce non-polynomial nonlinearities while preserving the symplectic structure of the problem and applied to the Vlasov-Poisson system.In the second part of the thesis, we consider several data-driven methods to address the poor accuracy in the under-resolved regime for Galerkin reduced models via a closure term. The closure term is developed systematically from the Mori-Zwanzig formalism by introducing projection operators on the spaces of resolved and unresolved scales, thus resulting in an additional memory integral term. The interaction between different scales turns out to be nonlocal in time and dominated by a high-dimensional orthogonal dynamics equation, which cannot be solved precisely and efficiently. Several classical methods in the field of statistical mechanics are used to approximate the memory term, exploiting the finiteness of the memory kernel support. We conclude this thesis by showing through numerical experiments how long short-term memory networks, i.e., machine learning structures characterized by feedback connections, represent a valid tool for approximating the additional memory term.

EPFL2022