5 74 5. Towards biologically plausible phosphene simulation offer advanced tools which allow to fit probabilistic retinotopic maps generated from large samples to any individual NHP brain (Klink et al., 2021; Messinger et al., 2021) and it is currently possible to accurately predict human cortical receptive field mapping based on anatomical scans (Benson et al., 2012; 2014), or other retinotopic mapping strategies that do not rely on visual input (Bock et al., 2015; Goebel et al., 2022). This opens new doors for future research into the functionality of visual prostheses with limited visual field coverage. Thanks to the machine learning compatibility, the model can also be used for the pre-operative optimization of the implant placement and design (van Hoof, 2022). Threshold and brightness The results presented in Figures Figure5.2andFigure5.3indicate that our simulator closely models existing psychophysical data on the stimulation thresholds and phosphene brightness, for different electrical stimulation settings. Note that the effects found by (Fernández et al., 2021) (that were modeled by us) are consistent with findings by other studies, which report brighter phosphenes for higher stimulation strengths (Schmidt et al., 1996), and a lower stimulation efficiency (i.e., higher total charge thresholds) for longer pulse trains or higher pulse widths and frequencies (Niketeghad et al., 2020).While the simulator is developed for intracortical electrodes and stimulation amplitudes in the range of micro-amperes, there is a wide range of literature describing the use of ECoG electrodes which are placed on the cortical surface (e.g., Beauchamp et al., 2020; Bosking et al., 2017a; Brindley and Lewin, 1968; Dobelle and Mladejovsky, 1974; Girvin et al., 1979; Niketeghad et al., 2020; Winawer and Parvizi, 2016). These electrodes require higher currents to elicit phosphenes (in the range of milli-amperes), but the mechanisms underlying the generation of phosphenes are presumably similar to those of intracortical electrodes. The results fromFigureS4, suggest that the implemented thresholding model for intracortical stimulation also generalizes to surface stimulation. Moreover, our results are in line with other computational models for detection thresholds in the somatosensory cortex (Fridman et al., 2010; Kim et al., 2017). Our results indicate how a leaky integrator model and normally-distributed activation thresholds, provide a suitable approximation of the tissue activation in cortical prostheses. Note that alternative, more complex, models can possibly predict the psychometric data more accurately. However, most probably, this will entail a trade-off with the simplicity and modularity of the current simulator. Future research may further improve our understanding of the neural processing underlying the conscious perception of phosphenes, possibly borrowing insights from the domain of natural vision. More elaborate theories on this matter have been developed and tested in (van Vugt et al., 2018). More specific limitations and suggestions for future adaptations are discussed in sectionSection 5.4.3. Temporal dynamics The results presented in Figure 5.4 reveal that the model accounts for experimental data on the accommodation in response to repeated stimulation in time periods up to 200 seconds. However, in contrast to the findings by (Schmidt et al., 1996), our simulator predicts a moderate recovery over the next 1000 seconds. Although we cannot provide an explanation for this difference, the modelled recovery largely stays within the 95% confidence interval of the experimental data. Similar to the other components of our simulator, the memory trace was chosen as a basic, yet effective, model of neural
RkJQdWJsaXNoZXIy MTk4NDMw