In-vitro 3D organoid cultures constitute an essential component of modern-day biological research, with key application areas in drug and therapy testing, and Organ-on-a-Chip models. The increasing demand for reliable models comes with the challenge of boosting the throughput of production of these cultures. The currently researched 3D culture methods pose the challenge of not being contact-free, of limited versatility and throughput in organoid culturing, and often lack a stimulus to promote cell agglomeration. This work presents acoustofluidics as a stimulus-driven solution to promote the clustering of cells in a levitated, suspended environment. In order to establish this objective experimentally, an SBAW resonator was designed, fabricated, and tested as a PoC (Proof-of-Concept). The experiments were carried out with the fabricated PoC which was characterized to assess the trapping performance and determine the controllable experimental variables to tune the acoustofluidic effects aptly, leading to the formation of stable clusters in SBAW nodal planes. On successful fabrication, the device showed repeatable trapping over a wide bandwidth of 300 kHz, and trap stiffnesses of up to 1.7 fN/μm determined experimentally based on the processing of particle agglomerate imaging. It was observed that the trap stiffness was adequate for levitation over several hours, and yet allowed for the formation of a 3D agglomerate of particles. It was also shown that a frequency-sweep actuation was successful in overcoming fabrication limitations and suppressing streaming, and a suitable working range of experimental parameters could be determined to achieve initial clustering in under 3 minutes. A futuristic outlook on the in-plane particle confinement methods to further improve the target performance, and considerations for biological experiments as the immediate next step have been presented as a concluding note in this thesis. This work thus paves the way for the integration of this technique into laboratory organoid culture formats. This positively complements the objective of cutting down the agglomeration time for initial organoid clustering, for an anticipated positive impact on the production throughput and developmental aspects of organoid cultures, with the in-house fabricated and characterized PoC presented in this work as an enabler.

As technology is scaling down by reducing the lateral and vertical dimensions of transistors. It becomes necessary to reduce the supply voltage, enhancing lower power dissipation and to maintain device reliability. This scaling makes the digitally controlled oscillator (DCO) more susceptible to interference between systems on the chip (SoC). For instance, the coupling from the power amplifier(PA) leads to injection pulling of the DCO. Especially in the non-constant envelope modulation schemes, e.g. π/4-DQPSK and DQPSK, the frequency pulling is extremely troublesome causing severe disrupted output spectrum. Now that Bluetooth provides, besides the Bluetooth Basic Rate(BR) also Enhanced Data Rate (EDR) and Bluetooth Low Energy(BLE) a radio with multiple purposes becomes more complex. Especially in the case where EDR uses π/4-DQPSK or DQPSK and the BLE only GFSK modulation. Therefore, in this thesis a dual frequency DCO at 4.8GHz and 9.6GHz is designed to mitigate the coupling issues with the PA. It also includes a study in AM-FM, where an optimal amplitude of oscillation shows minimal variation of frequency. An overall low power is maintained at 1V supply and a current of 308μA and 765μA for BLE and EDR respectively. Moreover, the thesis shows an attempt to leave out the Low Dropout Regulator to address future scaling challenges ahead. The DCO is simulated in 22nm-FDSOI occupying a core area of 0.0543 mm2 and achieves a Phase Noise of -116 dBc/Hz after dividing the frequencies 5.06 GHz and 10.74 GHz into the 2.4 GHz band

Superconducting Nanowire Single-Photon Detectors (SNSPDs) are characterized by high detection efficiency, high counting rates, low dark count rates, and minimal timing jitter, making them indispensable in fields such as quantum information science, free-space optical communication, and fundamental physics. Conventional SNSPD architectures, however, require the superconducting nanowires to be biased with a constant current and necessitate continuously operating readout circuits to capture the output signals. This leads to efficiency issues under prolonged low-load conditions and poses significant challenges for the development of multi-pixel SNSPD arrays. This thesis begins by examining conventional SNSPD designs, critically analyzing the shortcomings of previous active quenching methods, and proposing improvements to the digital sub-circuit. The introduction of a differential amplifier in place of the original single-ended main amplifier successfully addresses output offset issues, pushing the count rate beyond 50MHz. Moreover, this work introduces a novel persistent current SNSPD that leverages the memory characteristic of persistent currents in a superconducting loop, enabling photon detection without the continuous drive from interface circuits, and allowing for unified readout. A design methodology for SNSPD loops is proposed based on various simulation techniques and the physical and electrical characteristics of SNSPDs. This methodology was applied to design multiple samples tailored to different loop coupling scenarios. Additionally, the thesis outlines the biasing and readout logic for the persistent current SNSPD, ensuring independent photon detection without external circuit interference. In the demonstration system, the energy consumption per detection event was measured at 7pJ, with the average power consumption dependent on the frequency of rebias operations.

Abstract—This paper investigates the performance of commonly used spike detection algorithms (Absolute Amplitude Thresholding and Non-linear Energy Operator) on compressed neural signals using a novel wired-OR lossy compression algorithm. Performing compression with the wired-OR architecture mainly removes the noisy baseline and preserves spikes in the neural signal. As a result, the spike detection sensitivity and accuracy improve or stay similar after compression. In addition, this paper proposes a new spike detection algorithm, the non-zero spike detector, that can be efficiently integrated into hardware with the wired-OR compression scheme. By using a firing rate-based approach to optimize the threshold in the non-zero spike detector, the proposed technique outperforms both Absolute Amplitude Thresholding and the Non-linear Energy Operator across different signal-to-noise ratios and firing rates. The wired-OR readout architecture, in combination with the non-zero spike detector, is a promising approach to achieve massive compression while preserving the neural signal and maintaining spike detection performance.

3D-multi-electrode arrays (3D-MEAs) are needed to overcome the limitations of 2D-multi-electrode arrays (2D-MEAs) and enable the electrical characterization of 3D neuronal cultures in in vitro brain models, advancing the understanding of neurological disorders and paving the way to personalized medicine. The aim of this thesis was to overcome some of the limitations of current 3D-MEA devices and develop structures approaching the stiffness of the brain microenvironment, by using materials softer than conventional Silicon.A polymeric 3D-MEA was designed and developed by means of an innovative combination of two-photon polymerization (2PP), a 3D printing technology with sub-micrometer resolution, and standard wafer-level microfabrication methods from the semiconductor industry. Two novel fabrication protocols were developed, the first being a combination of 2PP with high-aspect ratio photolithography, which, though feasible, proved to require an inconveniently laborious process flow. The second fabrication process employed instead 2PP to fabricate the polymeric structures, pattern the microelectrodes, and provide electrical insulation. The 2PP-based process flow was ultimately preferred due to its potential for fabrication of structures of higher aspect ratio and geometrical complexity for 3D-MEA, extending their measurement resolution. Furthermore, a wafer-level alignment routine was developed with an alignment repeatability of 2PP structures of ±5 µm, which enabled the multi-step 2PP fabrication process. A novel maskless photolithography via 2PP process was also developed to pattern thin films over slanted surfaces, utilizing photoresist and glycerol-based immersion optics. The resulting 3D-MEA consisted of 15 printed polymeric pyramids featuring a total of 60 gold microelectrodes. The electrical insulation of the traces was partially successful, and will require further process development.The results demonstrate the feasibility of merging, for the first time, the 2PP process with standard wafer-level microfabrication techniques, specifically for the fabrication of a 3D-MEA for in vitro studies of human induced pluripotent stem cell (hiPSC) neuronal cultures. The 2PP-based solutions provided in this thesis show a promising pathway for the development of more complex and biomimetic 3D-MEAs. More generally, the developed wafer-level alignment routine and maskless photolithography via 2PP process for high-aspect ratio structures contribute to advance the field of microfabrication, and may enable the development of other types of innovative microdevices.

Class D amplifiers (CDA) have been used considerably in audio applications due to their high-efficiency behavior. Compared to standalone analog input CDAs, digital input CDAs are less sensitive to electromagnetic interference (EMI), which is preferred in automotive applications. For digital input CDAs, multi-bit current-steering DACs using tri-level units offer high dynamic range (DR) compared with other types of DAC. The main challenge in designing low total harmonic distortion plus noise (THD+N) and high DR IDAC lies in noise, mismatch, and ISI reduction. In this project, a new ISI resilient dynamic element matching technique is developed for mismatch shaping and ISI reduction in IDAC.

For the heart to pump blood throughout the body, electrical impulses that trigger the cellular contraction must be generated and spread through the myocardial tissue. These signals propagate faster along the longitudinal cardiac fiber direction than the transverse direction, conferring the heart with anisotropic conduction properties. Therefore, the arrangement of the fibers within the tissue governs the impulse propagation. Given the variability of the fiber direction across the heart and between patients, incorporating it into electrophysiological models would enhance our understanding of the mechanisms and progression of different heart conditions, such as atrial fibrillation (AF). The study of this common cardiac arrhythmia relies on analyzing electrical recordings of the heart, known as electrograms (EGMs), which, if integrated with the patient’s fiber architecture into cardiac models, can enable effective personalized treatment. Over the years, researchers have proposed different approaches to estimate the fiber direction from EGMs. However, these methods have been evaluated in different, usually simplistic, cardiac tissue models, making their comparison, and therefore selection of the most accurate approach for clinical and research applications, challenging.

The current study aims to identify the best fiber direction estimation method under consistent and realistic conditions. To achieve this goal, synthetic EGMs and local activation time (LAT) maps were generated from 2D and 3D monodomain models that mimicked the muscle bundle, atrial bilayer, and ventricular transmural fiber rotation structures. A comparison analysis of existing fiber direction estimation methods, first as described by their authors and then standardized to have the same spatial resolution, showed the superior performance of the techniques based on fitting an ellipse to local conduction velocity or conduction slowness vectors from a whole LAT map. The estimation accuracy of these methods can be further improved by increasing the number of vectors to which the ellipse is fitted. Nonetheless, given the influence of underlying layers in the epicardial recordings, the estimation error increases in the tissue models where fibers in the epicardial and endocardial layers run perpendicularly. The effect on the estimate of such architecture, characteristic of the inferior side of the right atria and the ventricles, can be accounted for by combining epicardial electrical recordings obtained after pacing either in the endocardium or the epicardium. Although a preliminary assessment of the estimation methods was carried out with human EGMs, future studies should focus on validating the methods in a controlled experimental framework and refining them for more localized fiber direction estimation. All in all, the automation of the techniques and their integration into electrophysiological models brings us a step closer to creating valuable clinical tools for diagnosing and treating electropathologies.

Traditional power batteries have been unable to meet the requirements of wireless sensor network (WSN) applications which require battery lifetime and no battery replacement. So, harvesting ambient energy and converting it into electrical energy to power wireless sensors has become an alternative to the traditional power sources. Harvesting energy from heat using thermoelectric generators (TEG) has become a prevalent topic in recent years. Since the system is required to startup with zero energy stored, the cold startup technique is necessary to activate the operation of the system in the very beginning stage without any assistance from the energy storage block, for instance, battery and capacitor. It is noted that any machines that moves, shakes, or rotates not only dissipate heat while working but also vibrate, thus the mechanical vibration can be used to startup the system. In this thesis project, a novel piezoelectric generator (PEG) startup technique is proposed to activate the system. This project focuses on pushing the minimum startup TEG voltage to an ultra-low level with the help of the PEG startup circuit.

The design of the system follows the top-down approach. Firstly, the inductive DC/DC boost converter was chosen to be the basic topology. Then, the Maximum Power Point Tracking (MPPT) technique is exploited to make sure the TEG always works at its optimal state. In addition, a Zero-current Switching (ZCS) block is proposed to regulate the synchronization of the switches and control the gate switching. Next, the state detector block monitors the state of the harvester system and instructs other control circuits to respond to the changes in the operation state. To avoid the impact of the fluctuation of supply on powering internal control circuits, a low-dropout regulator (LDO) is adopted to generate a stable supply instead. Finally, the power losses, conversion efficiency and startup principle are analyzed and calculated for performance improvement.

The need to understand how the brain works has sparked interest in electrode arrays to record neural signals. Intracellular recordings with nanoelectrodes that penetrate the cell allow better signal quality and additional functions such as voltage clamping to set the neuron's transmembrane voltage. The voltage clamp aims to facilitate the independent study of the ionic channel contributions to the overall neuron's transmembrane current for applications such as drug screening. This thesis aims to design a voltage clamp with CMOS technology to integrate into an intracellular neural interface system. The circuit, a trans-impedance amplifier, needs to be low-noise, small, and have wide clamping and output voltage ranges. The amplifier's design follows a structured methodology and uses symbolic analysis with the software SLiCAP and simplified EKV models aiming for easier readjustments and transparency for other researchers. The controller is an Operational Amplifier with a differential pair input-stage, an unbalanced common-source output-stage, and frequency compensation. The feedback network is a pseudo-resistor due to the enormous gains required to record the currents in the electrode of about 100 pA - 1 nA. The final simulations estimate a trans-impedance gain from the electrode to the load range of [2 M,1 G] V/A, a clamping voltage range of [0.90, 3.16] V, an output voltage range of [0.50, 3.16] V, and an output-referred noise between 0.095 and 0.41 mV_rms depending on the gain. The power usage is 49.1 uW/pixel and the area 0.0092 mm²/pixel. Post-layout simulations show stability degradation due to the influence of parasitics in the pseudo-resistor.

Non-contact infrared thermal imaging provides a non-invasive, safe and accurate means for screening subjects. COVID-19 (a disease caused by the Severe Acute Respiratory Syndrome Corona Virus 2 (SARS-CoV-2)) has changed how health care workers (HCW) interact with patients. Despite the usage of personal protective equipment (PPE), social distancing protocols, and unique facilities to screen individuals, the increase of infection rates among HCW is high. This has enormous consequences as HCW cannot provide direct patient care, thereby decreasing the workforce’s availability during the pandemic and has fuelled the urgent need for a non-invasive IR system to screen for potential cases of infections. Existing remote screening systems are susceptible to shortcomings caused by location and technique used for measurement, distance-ambient temperature associated inaccuracies and inefficient integration of hardware and software modules, invalidating their utility in the current situation. During this project, a non-invasive screening prototype was developed that mitigates these shortcomings by using an adaptive measurement method with automatic environmental compensation for screening elevated skin temperature and respiratory rate. The prototype is tested and validated in a controlled environment.

Disruptive changes in the fields ofcryptography and quantum chemistry can be achieved through quantum computing,as specific computations can be sped up significantly. Thousands to millions ofqubits are required to perform these computations, illustrating the need forlarge quantum computers. These qubits must be cooled to cryogenic temperaturesin order to operate. Currently operations on few qubits can be done, these arecommonly controlled with room temperature equipment. However as the number ofqubits becomes larger, scalable and integrated electronics is required toreduce the interconnect. This has promoted research in cryogenic electronicsthat can be located close to the qubits. This thesis presents a low power,scalable cryogenic DC biasing solution for the biasing of spin qubits. Thepresented DC-DAC has been implemented in the Intel 22-nm finFET process. Unlikeconventional DACs, the designed DC-DAC periodically generates all possiblevoltages in the full scale output voltage range, and drives a multiplexer thatconnects a sample and hold capacitors when the right bias voltage for theelectrode has been generated. The DAC is able to achieve a high resolution of 16bits with low noise performance by using an offset-compensated switched capacitorintegrator. By offering a DC input to the integrator the full scale outputvoltage range can be generated, while using the offset-compensated switchedcapacitor integrator allows reduction of offset and 1/f noise which are expectedto degrade at cryogenic temperatures. The DC-DAC is designed to meet a largevoltage output range of 3 V, which breaks compliance with the nominal 1.8 Vcompliance of the thick oxide transistors in the process. While it is possibleto put larger voltages over the transistors, it is currently unknown how thiswill affect the reliability and behavior of the thick oxide transistors.Therefore, a two-stage amplifier has been implemented where the second stageuses a supply of 3.6 V. The voltage compliance in the second stage isguaranteed through cascode transistors that are dynamically biased byadditional DACs. As the addition of the second stage of the amplifier andcoarse DACs introduce another voltage domain, additional circuitry such aslogic level shifters, diode stacks and cascode transistors have been added toensure that voltage compliance is throughout the entire DAC. The DC-DACachieves a LSB step of 45.1휇푉,, avoltage output range of 3 푉,,integrated noise of 17 휇푉 rms and dissipates a total of 187휇푊, whilenot consuming more than 0.078 푚푚,2. Thehigh resolution, large voltage output range and low power of the DAC enableintegration of the DAC with the qubits, paving the way for scalable quantumcomputers.

Soil water-content sensors based on the measurement of soil impedance are widely used for the detection of rain-fall induced slope failure and other agriculture applications. In principle, low-leakage sensors can measure soil water content down to 4%, but they require readout systems capable of generating soil excitation signals at frequencies ranging from 1KHz to 1MHz. However, existing commercial products do not cover this frequency range. To solve this problem, a new readout system is proposed in this thesis. By using sinusoidal excitation, implemented by a pulse-width- modulated driver, and synchronous detection, the readout system achieves less than 1% measurement error from 1KHz to 1MHz.

Capacitive micromachined ultrasonic transducer (CMUT) technology has been considered as a promising alternative for the conventional piezo electric-based technology in ultrasound imaging systems. Its potential advantages include better image quality, higher operational frequency, and ease of fabrication and
integration with CMOS read-out circuitry. However, CMUTs usually need high-voltage DC bias in order to transmit and receive acoustic waves. This work presents the design of a DC-DC converter to bias a CMUT for ultrasound imaging systems. Two different cases have been investigated: biasing at 63 V from an input of 5 V using a fully-integrated converter realized in a 180nm BCD technology that can handle up to 65V; and biasing at 120 V by employing minimal off-chip components. The proposed work explores the transient properties of a conventional on-chip switched capacitor DC-DC converter: the Dickson charge pump. A MATLAB model of the Dickson charge pump is developed to understand the relation between dynamic efficiency and the charge pump capacitor. Prior research works suggest that the dynamic efficiency of the Dickson charge pump is limited for a particular number of stages. However, the results obtained from the MATLAB model suggests that the dynamic efficiency can be improved by making the charge pump capacitor smaller. Various circuit simulations have been done to understand the results obtained from the MATLAB model. When the charge pump capacitor becomes very small, the parasitics start dominating and affect the overall efficiency of the Dickson charge pump. An optimization strategy is discussed to find the optimum number of stages and charge pump capacitor value to maximize the dynamic efficiency and minimize the circuit area. Based on the results obtained from this study, a DC-DC converter is designed to bias the CMUT at 63V, which consumes an estimated circuit area of 0.6586mm2 and has a simulated efficiency of 0.752.
In addition, a new design idea that incorporates two DC-DC converters to bias the CMUT beyond the process voltage limitations is also discussed in this thesis. This hybrid converter involves the optimized DC DC converter designed earlier to generate 63V from an input of 5V, a control unit circuit, and some off-chip
components. The hybrid converter designed consumes an estimated circuit area of 0.842mm2 and has a simulated efficiency of 0.577

Development of a visual prosthesis has the potential to help millions of patients with visual impairment around the world. One approach of visual prostheses is intracortical stimulation, where electrodes penetrating the visual cortex are used to create small light dots in the patients visual field. Said therapy requires an implantable neurostimulator that is able to stimulate more than 1000 micro-electrodes for sufficient resolution. Conventional stimulation methods are not suitable for this purpose as they are inefficient in multi-channel stimulation or lack control of injected charge. In this thesis a novel stimulation method is presented based on dynamic duty cycle control of ultrahigh frequency (UHF) pulses. The duty cycle is charge-controlled using a monostable multivibrator control loop. With this novel topology, the injected charge of a stimulation pulse is controlled even when electrode impedance changes. The presented system uses spatio-temporal stimulation to share the stimulator circuit among multiple output channels. Another contribution of this work is the implementation of an active charge balancing method. This method monitors the residual voltage at the electrode-tissue interface in between UHF pulses and stops stimulation when this voltage is brought back close to zero. The charge balancing circuit consists of a single comparator connected to the stimulation path. Overall, the proposed stimulator circuit consists of two comparators, one capacitor, three logic gates, and switches. The circuit is easily scalable depending on the stimulation parameters, requiring only two switches per output electrode. For validation purposes, the circuit has been implemented on a printed circuit board (PCB). The PCB has 8 output pins and operates up to 15V. Using a linear model of the electrode tissue interface, charge injection accuracy of the circuit was measured. The charge injected during a single UHF pulse is scalable with C*V. The implemented PCB uses a capacitor value of C=400pF, while the voltage V is scaled from 62.5mV to 1.25V such that the stimulation intensity ranges from 25pC to 500pC perUHF pulse. Furthermore, the charge balancing method was verified with the linear tissue model. The proposed method successfully reduced the residual voltage to 3.1mV, well within the safety limit of50mV. Measurements on a micro-electrode array confirmed functionality of the circuit for non-linear loads. Finally, the work presents an important insight regarding the development of power efficient neurostimulators. It has been shown that it is important to consider not only power efficiency of the circuit but also the energy efficiency of the stimulation waveform to decrease the power consumption of the system.

This work presents a type of low noise amplifiers (LNA) that are used for ultrasound imaging systems. To account for the attenuating nature of ultrasound echo signals, a so-called time gain compensation (TGC) circuit is required. By increasing the gain over time, the output dynamic range is decreased, while the switching artifacts are suppressed by the continuous gain control.
The proposed work combines the LNA structure with TGC functionality. This is done, such that the first component in the ultrasound receive chain does not need to be able to handle the full dynamic range of the input signal. The goal is to reduce die area and power consumption costs compared to systems that utilize separate LNAs and TGCs. The combined TGC-LNA consists of a transimpedance amplifier (TIA) with an exponentially-varying feedback resistance. As the feedback network and feed-forward path are separately designed, the design of the TGC functionality and low noise functionality can for a large part be independently designed. The TGC functionality is implemented by an exponentially-varying feedback resistance. This is achieved by means of implementing triode transistors as voltage-controlled resistors. Three branches with differently sized triode devices are required to obtain the full gain range of 40 dB. A two-stage telescopic amplifier realizes the loop amplifier. The bias current of the first stage is a linear function of the total feedback resistance to create a constant unity-gain bandwidth in a power efficient method. Realized in 0.180 μm BCDMOS technology, the combined TGC-LNA amplifies the signals from a Capacitive Micromachined Ultrasound Transducer (CMUT) with a center frequency of 7.5 MHz. The total achieved gain range is 40 dB where the gain varies during a receive period of 100 μs. During the receive period the total harmonic distortion remains below -44 dB and the noise floor is 1.12 pA/√(Hz) at the highest gain setting. Drawing 5.5 mW from a 1.8 V supply and requiring approximately 0.01 mm² die area, the proposed TGC-LNA provides a new method for reducing power consumption and area costs for miniature ultrasound applications.

Approximately 15-20% of theworld population is aected by tinnitus, a hearing condition associated with phantomsound perception. A large number of suerers experience a severe level of tinnitus and they are not ableto conduct a normal life because they develop insomnia, depression, and distress.Extensive research has been made to study its pathophysiology and to nd a therapy. However, currently thereare no treatments that have demonstrated to be eective in modulating tinnitus or suppressing its related annoyancefrom a long-term perspective. What is known is that the patients exhibit abnormal electrical activityin multiple areas of the auditory and the central nervous systems. This is the reason why the focus of thescientists has shifted to multi-modal stimulation: multiple stimulations of equal or dierent nature (e.g. doubleelectrical stimulation or acoustic-electrical stimulation) are applied at the same time in the attempt to induceneuroplasticity and restore the normal electrical activity in the targeted areas. Multi-modal stimulation hasbrought signicant improvements both in terms of tinnitus intensity and distress, but the studies conductedare too little to derive conclusions. Stimulation sites, parameters, and patterns are some of the many issues thathave to be properly investigated.The challenge of this work is to design a portable multi-modal stimulator that is able to provide bilateral acousticand electrical stimulation simultaneously. The device works at the same time as an audio player and atranscutaneous electrical nerve stimulator, with the purpose of contemporaneously stimulating the auditorycortex and the autonomic nervous system or the dorsal cochlear nucleus through the vagal nerve or the C2nerve, respectively.Bi-modal stimulation is based on the random presentation of pure tones matched to the tinnitus frequency pitchcombined with two possible electrical stimulation waveforms: a novel one characterized by the superposition ofa low-frequency noise on a DC component ("noise + DC" stimulation) and the second one is burst stimulation.Both the electrical stimulations are current-driven and largely customizable due to the wide programmabilityof the stimulation parameters. The analog design is realized in such a way that two identical output currentsare delivered to the tissue, allowing for bilateral stimulation.The low power consumption and the small dimensions and weight of the device will permit the tinnitus patientto use it for several hours per day while performing his daily life without impairments.

Data analysis for electrical impedance tomography (EIT) research requires manual selection of sequences with a normal breathing pattern. This procedure is lengthy and the lack of a standardised approach results in different practices among EIT studies, limiting the potential and comparability of the research.
This article presents a new approach to solving this problem, using automatic detection of EIT sequences with normal breathing pattern. An algorithm was developed to differentiate between normal and disturbed breathing patterns. To facilitate data analysis, it was implemented in an application that allows for EIT parameter calculation of selected sequences. EIT recordings of three patients recruited in an observational study were used to develop the algorithm. A reference standard was defined as the majority vote of five biomechanical engineering students performing manual classification. Frequency and time domain properties were compared between the signals that these graders classified as reliable and unreliable, and were used to define classification rules for the algorithm. Matlab was used to create the classification algorithm and implement it in an application. The developed algorithm was validated with a new data set containing EIT recordings of an additional three patients, classified by the same volunteers. Qualitative analysis was performed to investigate the causes of conflict between manual and automated data selection.
The resulting algorithm achieved a sensitivity of 92.8% (95% CI, 92.6%-92.9%) and a specificity of 85.5% (95% CI, 85.1%-85.8%) on the EIT files used for development. On the validation set the algorithm accomplished a sensitivity of 86.5% (95% CI, 86.3%-86.7%), and specificity of 79.7% (95% CI, 79.4%-80.0%).
Most differences between manual and automatic data selection were found to be around the edges of the selected sequences. Other discrepancies can be explained by difference in data selection behaviour for varying recording qualities during manual selection. The presented algorithm proves its potential for quick and reliable data classification of EIT recordings. It not only provides a new standard for data selection in EIT research, but also reduces the time investment of researchers. The freely available data analysis application enables easy implementation of the algorithm. The presented study therefore provides a first step towards a uniform approach in EIT research, improving comparability of studies and increasing the scientific value of their findings.