Epilepsy is a neurological disorder that affects millions of people worldwide and is characterized by recurrent seizures. Managing epilepsy effectively remains a challenge, particularly for patients who do not respond to medication. Closed-loop neuromodulation systems have emerged as a promising approach for real-time seizure detection and intervention, offering the potential to improve the quality of life for individuals with epilepsy. These systems monitor brain activity and deliver therapeutic stimuli when abnormal patterns, indicative of an impending seizure, are detected. However, the effectiveness of these systems depends heavily on the accuracy and efficiency of the embedded seizure detection algorithms.

Implementing brain-computer interfaces (BCIs) for neuromodulation presents significant challenges, particularly in terms of power consumption, area constraints, and computational complexity. The use of machine learning (ML), especially lightweight classifiers, is critical for overcoming these challenges. ML models can be trained to identify complex seizure patterns with high sensitivity and specificity, but their deployment in implantable devices requires careful consideration of hardware limitations.

This thesis addresses these challenges by employing an oblique tree estimator to develop a low-power, low-area seizure detection classifier that relies solely on temporal features to minimize hardware costs. Specifically, the Sparse Projection Oblique Randomer Forest (SPORF) variant is utilized, to enable multiplier-less classifications, along with a multi-path inference approach thereby reducing memory requirements. The proposed model achieves state-of-the-art performance in both patient-specific (98.78%/ 99.89% and 98.52%/99.91% sensitivity/specificity in the CHB-MIT and ETHZ databases, respectively) and patient-independent (92.26%/98.29% in the CHB-MIT database) seizure detection frameworks. Additionally, post-layout simulations of the synthesized design reveal a power consumption of 1.22 µW and an area of 0.0076 mm², with an energy consumption of 7.58 nJ per classification using 40 nm CMOS technology.

Subjective) tinnitus is a phantom perception, where a sound is perceived by the subject that cannot be objectively perceived by other observers. an estimated 9.8% of people suffer from chronic tinnitus, moreover, for 1%-3.1% of people, it severely impacts their lives, leading to anxiety, depression, and insomnia, amongst other symptoms.
Tinnitus is strongly related to hearing loss, however, the ear is likely not the sole cause of tinnitus.

Currently, the generation and progression of tinnitus is thought to be a complex multifactorial process. Although research is ongoing and a consensus is not reached yet, the most recent theories consider the following mechanism for the development of tinnitus: (1) a lack of auditory input, triggers (2) over-expression in auditory neurons in the brainstem and auditory cortex, both part of the ascending pathway, where after (3) the tinnitus signal becomes wrongly labeled as important by the brain’s medial pathway and directed to the conscious parts of the brain where it is perceived, subsequently (4) the descending pathway is unsuccessful in suppressing the generation of tinnitus.

It is hypothesized that pairing acoustic stimulation using tinnitus-matched sound with transcutaneous electrical nerve stimulation of the vagus nerve can teach the brain to associate the distressing tinnitus sound with the de-stressing stimulation of the vagus nerve. Over time it is expected that the tinnitus sound will be labeled as unimportant by the medial pathway and is thus consequentially suppressed.

To investigate this hypothesis a bimodal stimulator is required that can combine acoustic stimulation and transcutaneous electrical vagus nerve stimulation. The device should be easy to use and safe. The specific focus of this thesis report was to design the electronic hardware for electrical stimulation and power management and implement various safety features.

From testing the stimulation output it could be concluded that the device is successfully able to produce the electrical pulse pattern and potentially any other electrical pattern. Following a risk assessment, and analyzing the safety of the device, it was concluded that the device can be used to test the efficacy of the proposed treatment, given that it is used in the prescribed circumstances and that the user is aware of the risks. Furthermore, it was shown that the device was able to stimulate the vagus nerve. Overall, the device was easy to use and was able to pair acoustic and electrical stimulation.

Background Electroencephalography (EEG) using a dry electrode cap is currently being investigated as a pre-hospital stroke triage instrument. Developing an algorithm for automatic interpretation of the EEG signals is challenging, considering the amount of artefacts often in the signal. Ideally, an algorithm should be capable of distinguishing between different artefact types to determine the appropriate action: whether to correct them, reject them, or consider their potential predictive value. Neural networks have demonstrated their value for these types of classifications in wet EEG data. However, this approach requires enormous amount of data and dry EEG data is sparse.
Objective This study aims to develop a multi-class artefact classification model for dry EEG data using transfer learning.
Methods First, a convolutional neural network (CNN) for multi-class (clean, eye movement, muscle activity and electrode artefact) classification was developed. Wet electrode EEG recordings from a publicly available dataset were used, containing data of 213 patients (Part I). Second, this model was implemented and transfer learned for multi-class (clean, pulse artefact, muscle activity and artefact) classification of dry EEG recordings using data of 13 subjects (Part II). The models were trained using annotated multi-channel input. Model performances were evaluated on unseen test data using accuracy, area under the receiver operating characteristic curve, F1-score, precision, and recall. 
Results The pre-trained multi-class model achieved an overall accuracy of 74.8%. The fine-tuned model was able to correctly differentiate between the classes with an accuracy of 71.2%, with the best performance for the classes muscle activity (AUC 0.92, F1-score 0.80) and artefact (AUC 0.94, F1-score 0.80). 
Conclusion Transfer learning enabled the development of a good performing multi-class artefact classification model specifically tailored for dry EEG data, even though available data was limited. The developed model could assist in assessing appropriate action for different artefact types in dry EEG data interpretation algorithms.

Purkinje cell is a type of neuron that can be found in the cerebellum. What characterises Purkinje cell neural activity is the fact that it exhibits two types of spiking behaviour; the so-called simple and complex spikes. These two types of spikes are thought to play a role in motor functionality. In order to better understand the relationship between Purkinje cell neural activity and the motor-cortex, neuroscientists record such neural activity in mice. However, current experimental setups pose a challenge as they involve a wired connection between the animal’s head stage and the recording device, which limits the mouse’s natural behaviour by restricting its movement. This work proposes a lightweight neural-spike detection and classification architecture for acquiring Purkinje cell neural activity. The proposed design discards unneeded information, by detecting and classifying spikes in real-time. This type of compression enables data storage on a removable device in the head stage, freeing mice from wires. Its small formfactor allows unrestricted movement during experiments, while a power-efficient design ensures long-termoperation. The performance of the algorithm has been evaluated using a software implementation, yielding a combined accuracy for detection and classification ranging from 92.74% to 94.54%. The system has been synthesised using the 45 nm Nangate Open Cell library resulting in an ASIC with an area of 0.22mm2 and a power consumption of 0.412mW.

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.

Implantable Medical Devices (IMDs) are deployed in patients to treat a range of medical conditions. Technological advancements have enabled manufacturers to fit IMDs with specialized hardware that accelerates compute-intensive medical therapies next to a software-run host processor. However, mostly hardware acceleration is found in the form of ASIC peripherals next to a host processor in state-of-the-art IMDs, while low-power FPGAs could provide a comparable performance gain with the added benefit of the upgradability of functionality. Existing literature about low-power FPGAs focuses on new algorithms or performance improvements, while largely ignoring power and energy analysis, the latter being the most limiting factor in the IMD environment. This thesis investigates under what conditions FPGAs could be added to IMDs by developing two use cases: an FPGA securing wireless communication, and accelerating a neural network aiding medical therapies that depend on pattern detection. These cases are evaluated on FPGA, eFPGA and MCU with ASIC peripheral platforms, from which performance, energy usage and prospected IMD battery life is derived. On one end, it was found that AES encryption used 4.4 times the energy of an MCU hardware-accelerated implementation while being 17% slower. However, employing lightweight ciphers on the FPGA closes this gap. Furthermore, adding an FPGA results in only a 7.5% decrease in battery life when the FPGA is shut off during idling to combat its high static current draw. Running an FPGA-accelerated neural network is feasible if the active time is 6.5 minutes per day. With weekly recharging, continuous monitoring is possible. Using an eFPGA, which is an embedded FPGA fabric integrated within an MCU, results in using only 12% to 21% of the FPGA package area and is almost 2 times as energy efficient under 2 minutes daily usage as an FPGA. As FPGAs in IMDs is a novel field, research was done in legal regulation of IMDs, where it was found that existing regulations on software devices also applies to FPGAs. Therefore, all obstacles of the technical and legal kind have been removed that hold IMD manufacturers from using FPGAs in their devices.

This project aims to explore the possibility of monitoring the vitals of the neonate remotely and without placing any sensors or electrodes on the subjects body using Thermal Imaging.

The field of active implantable medical devices (IMDs) has made huge steps forward in the last years. IMDs are used today for the treatment of previously untreatable conditions as well as the continuous monitoring of the health of patients. The field, where active IMDs have been the most successful, is the field of neurological conditions, on which traditional medicine such as the intake of pills has very limited effect. In order to take full advantage of the possibilities offered by the IMDs it is vital to have communication between them and the external world. In this way, the treatment method provided by the IMD can be configured and optimized for each separate individual, achieving the best possible results. Moreover, the state of the device itself can be assessed and therefore possible flaws in its operation can be detected and corrected. Finally, in case of monitoring the patient’s health condition communication is also vital.
However, this communication between the external world and the implanted device also introduces the possibility for someone, that possesses the required expertise and skills, to hack the device. The hacker can possibly “eavesdrop” the communication channel in order to extract information regarding the patient or even act in order to change parameters of the device and the therapy it provides, possibly causing harm to the patient. One way for hackers to disrupt the normal operation of the IMD is to drain its power source, namely the battery of the device, an attack that is named battery denial of service (BDoS) attack. One of the easiest ways to achieve such an attack is to make the device commit valuable power resources in order to run the authentication protocols continuously, by non-stop requesting for access. In this thesis, we try to tackle this problem by creating an analog circuit module that can act as an add-on to commercial IMDs that will have the task of authenticating the user that is trying to communicate with the IMD by making use of harvested wireless energy that is required for the communication. In this way, we expect to relieve the IMDs from spending time and resources on the authentication process. To that end, we designed and created an analog experimental prototype that employs both wireless power transfer and data communication, by simply making use of off-the-shelf components. The goal of this experimental prototype is to identify further challenges that can arise in a so called zero power defence scenario.
To that end we implement a system that makes use of near-field resonant inductive coupling as a wireless power transfer method and is able to provide some mW of power across distances of up to 2cm, an amount of power that is sufficient to drive a microcontroller that can execute a low power security protocol in order to authenticate the entity trying to access the receiver. In addition to that, it can achieve a downlink data rate of 200kbit/s using amplitude shift keying modulation and an uplink data rate of 1kbit/s by making use of passive load shift keying modulation. This system is a unique implementation, since it constitutes the only solution that implements a highly asymmetrical bidirectional link by making use of ASK and LSK modulation and has a completely wirelessly powered receiver. Moreover, since this implementation was achieved with simple off-the-shelf components, there is a large margin of improvement in terms of size and efficiency if such a system was to be designed as an ASIC.

Proper security mechanism are a crucial part of safe usage of Implantable Medical Devices. Multiple researchers presented various solutions to address this problem, basing them on different underlying principles. Within the scope of this thesis we perform a security analysis of the chosen authentication protocols. What is more, we present a new attack on a scheme based on physiological signal processing using a fuzzy vault cryptographic primitive. We exploit the fact that the signal generated by the heart beats does not change sufficiently in the frequency domain. Therefore it is possible that the adversary reuses signal recorded at some earlier point of time to authenticate to the implant in real time. We show in an experimental way that it is able to break the scheme with probability reaching 75%. Finally, we propose a novel lightweight authentication protocol based on hash chains. To ensure the applicability of our work, we have decided to use only energy efficient solutions, that is hash functions and block ciphers. In contrast to existing work, we have extended the threat model and considered the implant reader distrusted. We present a set of energy measurements to provide advantages of different elements to be used during implementation of our solution.

The field of Computing has been a significant catalyst for innovation across various segments of our lives. Computational neuroscience keeps demanding increased perfor- mance to implement powerful simulators able to closely approximate brain behavior using complex mathematical models. This resulted in various High-Performance Com- puting systems able accelerate the above simulation workloads. One of the challenges is how these applications are being ported to massively parallel accelerators that requires significant time and effort for designing and debugging. This thesis primary task is to optimize an existing hardware library for neural simulation. The above library uses one of the most widely used biophysically-meaningful neuron models called Hodgkin-Huxley. The library optimizations will be performed while following a design methodology to accelerate applications on Maxeler’s Data-Flow Engines (DFEs). A DFE is an FPGA- based accelerator incorporating a top-of-the-line reconfigurable device surrounded by high bandwidth, large capacity on-card memory. This work focused in the fully ex- tended model that had room for performance improvements. The result, an optimized model that takes advantage of the FPGA capabilities and achieve up to 2.66x speed up over the previous implementation. They key to this speedup is the use of fixed-point arithmetic that provides 2x speed up compared to the optimized floating-point version. Additionally, the model is implemented in multiple kernels in such a way that can be scaled up using multiple DFEs to achieve even greater performance.

Understanding the hidden organizational principles existing in the human brain was always one of the great challenges in Neuroscience. To uncover the way the brain functions, advancements in the fields of Medical Imaging and Computational Science have been of great importance. Powerful imaging tools, such as functional magnetic resonance imaging (fMRI) and positron emission tomography (PET), have already enabled scanning the whole brain volume and visualizing the brain functioning, both at rest and during task execution, to a significant degree. However, several limitations especially in spatiotemporal resolution led to the need for further advancements in the field of functional imaging. An alternative technique, that overcomes most of the previously existing problems, is functional ultrasound (fUS). fUS is capable of imaging even the microvasculature blood-flow dynamics in response to brain activation with high spatiotemporal resolution. The wealth of fUS-acquired data calls for advanced data-analytic methods to uncover new information, beyond the well-applied simple univariant correlation method. This is the main goal of this MSc thesis, to use a proper analysis technique, mainly borrowed from the same-principle fMRI technique, in order to produce powerful inferences. For this reason, a detailed literature review regarding fUS imaging and fMRI analysis methods is introduced. Then, the main analysis part is focused on the Independent Component Analysis (ICA) method, trying to segregate the brain into spatially independent components that share a similar activity response. Here, the whole processing pipeline is established, describing all the necessary preprocessing steps along with ICA parameters and approaches (single- and group-ICA) using the ICASSO software package. As a post-processing step, functional images-to-Allen brain atlas registration is also performed in order to identify the different regions represented in the ICA-derived spatial components. The effectiveness of the methods is assessed based on the collected results on different datasets, obtained from 2D visual-stimulation as well as 3D resting-state experiments conducted on mice at the Neuroscience department of the Erasmus MC. As a conclusion, ICA was able to separate different anatomical and functional sub-networks. More specifically, from the visual-stimulation experiments, brain regions such as Lateral geniculate nucleus (LGN) that play a role in the visual pathway are identified, while from the resting-state the spatial continuity of different regions is confirmed.