Software and hardware co-design for AI driven neurostimulation optimization

Current commercially available implantable peripheral nerve stimulation solutions demonstrate low efficacy across all target clinical use cases. Main causes are technological limitations in terms of (1) selective activation of organ-specific fibers while avoiding fibers causing side-effects, (2) non-existent methods to reliably and rapidly extract anatomical information at the stimulation location of the target nerve, and (3) lack of close-loop systems capable of facilitating adaptive stimulation on the fly. While recent research on selective fiber activation shows promising results, including a recently introduced intermittent interferential stimulation method at imec, extracting anatomical information and using it within an optimization framework of a closed-loop neuromodulation system is a major challenge. To realize such a solution, one would need to design a distributed optimization system that combines a cooperative optimization framework, including resource-restricted optimization at the implant site combined with partial optimization at the remote hub(s) having less resource constraints.

The need to limit the processing power and memory (i.e., power consumption) would require minimum operation at the implant side focusing mainly on basic processing / compression of neural readouts, bi-directional communication with the external (wearable) unit and fast closed-loop control of stimulation parameters. Similarly, optimization step at the remote hub(s) would require complex processing of available real and simulated anatomical models, as well as real and simulated neural responses to a high-dimensional stimulation parameter space. Through low-latency communication, data captured by the implant needs to be rapidly interpreted such that the relevant information can be fed back to the implant to optimize the stimulation method and/or parameters.

The goal of the PhD assignment is to explore architectural and implementation choices of designing and developing such a distributed neuromodulation optimization system and implementing several strategies focusing on software (SW) and hardware (HW) co-design. An important aspect of the assignment is innovating in the field of application of artificial intelligence methods through SW/HW co-design and validation of the proposed methods and architectural choices. Available databases of neurophysiological recordings and anatomical data, as well as computational models and results of simulations, for selected neural populations will be used to drive the initial research phases. Furthermore, available optimization tools and HW/SW implementations of compression and data analysis (spiking neural networks) will be available at the start of the research. The candidate is expected to interact with the neuromodulation, neurotechnology and AI experts, as well as the experts in the field of implant design and pre-clinical validation. Evaluation would include testing the proposed methods and technologies in vitro but also in vivo using simple animal models such as earthworms, but also in large animal models such as pigs.

---

At imec, we have developed a neuromodulation framework capable of stimulating neural tissue in vivo and capturing direct neural response using an in-house (peripheral) nerve stimulation and sensing system. This system has been in use to explore novel stimulation paradigms in simple animal models such as earthworms, but also in large animal models such as pigs. Furthermore, we have enhanced existing computational model platforms to be able to explore complex stimulation methods and paradigms, utilizing precise anatomical information of peripheral nerves innervating different organs. All these tools and expert support will be available for the PhD candidate to explore and evaluate developed concepts and prototypes within the project.

The envisioned PhD topic is aligned with the imec neuro roadmap that has the ambition to develop and deploy adaptive closed-loop neuromodulation solutions through several ongoing and planned projects and activities. Within the internal Stargazer project, close-loop framework and optimization methods are explored to facilitate selective peripheral nerve stimulation through neural readout and stimulation. The optimization methods are being explored as a part of an ongoing NIH grant () and AAA Neuro project in terms of optimization methods and framework for online experimentation. Furthermore, the basic ASIC and system technology developed within the internal Ignite and Ignitor projects as well as EU funded projects SPARCLE (ongoing) and IMPHORIA (in preparation) aims at integrated circuits and neuromodulation systems that facilitate deployment of neuromodulation optimization solutions.