Biophysically accurate and machine learning-based surrogate models to optimize neuroprosthesis design and operation

Electrical stimulation of the nervous system has emerged as a promising assistive technology in case of many injuries and illnesses across various parts of the nervous system. In particular, the invasive neuromodulation of the peripheral nervous system seems to be a good trade-off between selectivity and invasiveness, and is thus a candidate to perform motor, sensory and autonomic function restoration. More and more sophisticated computational models of neuromodulation have been developed in the last fifty years, allowing the assessment of the expected performance of various neuroprosthetic devices and electrical stimulation protocols and the explanation of the fundamental mechanisms behind the success of neuromodulation interfaces. The complexity of such models, though, has prevented their use inside of true model-based optimization routines, since repeated computation of stimulation outcomes under different stimulation conditions requires a prohibitive amount of simulation time. The main contribution of the present work is the development of surrogate models to accelerate the evaluation of the neural variable of interest under given stimulation protocols, thus allowing the formulation of neuroprosthesis optimization frameworks. The possibility to perform model-based optimization, enabled by the use of surrogate models, leads to the need to personalize the models to the functional anatomy of the target subject. In a clinical setting, this needs to be done without any additional invasive procedure, which would negatively affect the patient's wellbeing. Here, frameworks for the determination of the functional organization of the target neural structures in the implanted patient are presented and characterized in silico, to accommodate both cases of acceptable and unacceptable off-target stimulation. The methods introduced in this work are modular and considerations on the development and use of these computational tools for different impairments, to target different parts of the nervous system and in different stages of research (fundamental, preclinical, and clinical) are outlined. The present thesis work paves the way for the development of model-based optimization of neuroprosthetic devices, with the promise to increase the effectiveness of electrical stimulation applications and at the same time reducing the amount of experimental data and resources needed to reinstate lost bodily functions in different cohorts of patients.