Application-Specific Design of Ultra-Low-Power Low-Noise Vagus Nerve Front-End Circuits for the Detection of Epileptic Seizures

Epilepsy is a neurological condition affecting about 70 million patients worldwide, one third of which do not respond to medication. Closed-loop vagus nerve stimulation (VNS) is a promising alternative treatment for patients suffering from drug-resistant epilepsy, improving their quality of life thanks to a reduction in seizure frequency and severity. This neuromodulation therapy has the potential to be more efficacious than its open-loop counterpart by delivering stimulation at the seizure onset. It however requires accurate seizure detection capabilities. For that task, several biomarkers can be used, but detecting seizures from the vagus nerve electroneurogram (VENG) offers a minimally invasive potential solution. The targeted closed-loop VNS system based on the VENG is powered by a battery and implanted in the patient’s chest. Hence, it should work with a power budget around 50 µW for both sensing and stimulation to ensure a 10-year lifetime, 10-15 µW of which can be allocated to the analog front-end (AFE) circuits used for sensing. Due to the low µV-range amplitude of the VENG signals captured by an extraneural electrode, the acquisition of the signals is sensitive to noise and interference. In general, the most stringent AFE specifications are (i) an ultra-low power budget, (ii) a low intrinsic noise in the acquisition circuits, and (iii) the need for the rejection of electromyogram (EMG) interference from nearby muscles activity. The acquisition of VENG signals is scarcely addressed in the literature and no published AFE design meets the specifications of that application. In this thesis, we propose methodologies and circuit techniques for the design of an integrated VENG-specific AFE tackling the aforementioned challenges. This work spans across several abstraction layers: (i) studying the noise-power trade-off of the low-noise amplifier (LNA) from device level to circuit level, (ii) designing and modeling a complete ultra-low-power (ULP) AFE, and (iii) linking circuit-level performance and non-idealities to seizure detection accuracy. The approach chosen in this thesis consists in relying on system and application knowledge to design and validate highly efficient circuits specific to VENG sensing. System knowledge includes the characteristics of the input signals and of the nerve interfacing electrode, and details about the seizure detection algorithms. The first chapter of this thesis investigates the context of this work, from biomedical application to circuit design. The choice of the VENG sensing application is motivated by medical needs and ethical concerns over other treatment solutions. Next, our knowledge about the closed-loop VNS system is linked to analog circuit specifications (e.g., noise and power consumption). State-of-the-art circuits are presented to meet those constraints in a VENG-specific AFE. Finally, seizure detection algorithms are presented to offer insights on the signal processing tasks performed after signal acquisition. The second chapter of this thesis dives into the study of the LNA, which is the bottleneck of the system in terms of noise and power consumption. Hence, the fundamental limits of the noise-power trade-off in biomedical amplifiers are analyzed with a semi-analytical approach using the gm/ID methodology and a limited number of technology-dependent parameters. We show that the minimum achievable noise efficiency factor (NEF) with a fixed topology can be accurately estimated analytically, allowing designers to compare LNA topologies without requiring time-consuming numerical optimizations. The novel methodology is then applied to the design and optimization of an LNA with an innovative topology combining current reuse for noise efficiency, and common-gate input for resistorless high-pass filtering of interference. The proposed LNA is prototyped on a 22-nm FD-SOI chip and reaches an NEF of 1.97, representing a 2× NEF improvement compared to the only existing VENG-specific LNA in the state of the art. In the third chapter of this work, the previously designed LNA is integrated into a complete low-power AFE dedicated to the acquisition of VENG signals. It is first demonstrated that rejecting in-band EMG requires a true tripole input with electrode imbalance compensation and low gain mismatch. While imbalance compensation was tackled in previous works, gain mismatch still has to be addressed. To do so, we propose an inter-channel chopping technique that performs dynamic element matching and modulates the residual interference to high frequency while keeping the VENG signal in its original band, achieving a measured gain mismatch level of 0.35%. The analog circuits are interfaced with ULP circuits including filters and an analog-to-digital converter to form a complete front-end. The proposed circuits are prototyped in a 65-nm CMOS technology and consume an average power of 13.3 µW, which is at the limit of the AFE's power budget. We show that the proposed front-end circuits enable the accurate capture of action potential waveforms with an amplitude down to 20 µVpp, which is the expected amplitude of VENG signals. The last chapter of this thesis bridges the gap between the circuit-level specifications of the front-end and the performance of the seizure detection algorithm. Bio-interface circuit specifications are often based on preconceptions of applicative requirements, and an arbitrary signal-to-noise ratio. However, the exact needs of the applicative context are unknown, leading to non-functional or inefficient designs due to too stringent specifications. This chapter uses experimental VENG data obtained on rats, a behavioral model of the front-end, and a seizure detection algorithm to study the sensitivity of detection accuracy to AFE limitations. The sensitivity analysis results are then leveraged to extract optimal specifications for the front-end while minimizing its power consumption. We show that it is possible to reduce the front-end power consumption down to 2.1 µW by relaxing its specifications while maintaining excellent seizure detection accuracy, paving the way for more efficient VENG-specific front-end designs and closed-loop VNS implants with a longer lifetime.

Affiliations

UCLouvainSST/ICTM/ELEN - Pôle en ingénierie électrique

Citations

APA
Chicago
FWB

Favresse, S. (2026). Application-Specific Design of Ultra-Low-Power Low-Noise Vagus Nerve Front-End Circuits for the Detection of Epileptic Seizures. https://hdl.handle.net/2078.5/281451