We are inherently sensorial beings. Our sensory organs continuously transmit environmental information to the brain, enabling us to perceive and interact with the world. When this transmission is disrupted-such as through traumatic injury and amputation-the resulting sensory loss can profoundly diminish quality of life. Neuroprosthetic systems aim to address this issue by interfacing with the nervous system to deliver sensory feedback. While progress has been made in auditory prostheses, restoring somatosensory and visual feedback remains a major challenge. For example, in somatosensory feedback, transcutaneous electrical nerve stimulation (TENS) is used to evoke tactile sensations, but it is often associated with unnatural percepts, typically described as tingling. Moreover,most current prosthetic systems focus on a single sensory modality, limiting the realism and richness of feedback.
This thesis is guided by two central research questions: how stimulation parameters can be optimized to evoke sensations perceived as more natural by the user, and how the sensory experience can be enriched by delivering multiple feedback modalities. The thesis first offers a conceptual perspective on various optimization frameworks that can be employed to refine stimulation strategies for eliciting naturalistic sensations. Two of these frameworks are subsequently implemented and evaluated.
The first is a self-optimization framework (SELFOPT), which enables users to iteratively adjust TENS parameters in real time. Based on the experimental results, participants using SELFOPT managed to find stimulation parameters with enhanced naturalness compared to conventional approach. Experimental findings also reveal a nuanced interplay between pulse amplitude, pulse width, and pulse frequency. While amplitude plays a dominant role in modulating perceived intensity, pulse width and frequency appear to fine-tune the qualitative aspects of the sensation, contributing to the increase in perception of naturalness.
The second framework explored is the actor-model framework, which demonstrates how the construction of a biologically digital twin can accelerate the optimization of stimulation parameters. Applied in the context of visual prosthetics, this framework was used to evaluate different image downsampling strategies. The model identified a downsampling method that produced neuronal responses more closely aligned with those elicited by high-resolution images, outperforming standard approaches. This result highlights the potential of actormodel framework to informpreprocessing algorithms for visual prosthesis applications.
To enrich the sensory experience beyond a single modality, the thesis introduces ThermalX-TENS, a multimodal device that delivers co-localized thermal and TENS feedback. Experimental validation showed that both modalities could be perceived simultaneously. Furthermore, real-time object discrimination tasks conducted with amputee participants demonst
