5 66 5. Towards biologically plausible phosphene simulation • InEquations (5.1) to (5.4) we usea=0.75, k=17.3, b=120, andα=0.95, based on a fitby (Polimeni et al., 2006) on data of the human V1 from (Horton, 1991). • InEquation(5.6), the parameter K is set to 675µA/mm2, following an estimate by (Tehovnik et al., 2006), who measured the responses to intracortical stimulation in V1 at different current levels. • InEquation(5.7), we use a rheobase current I0 =23.9µA based on a fit on data from(Fernández et al., 2021). Here, we used the strength-duration curve for tissueexcitabilityQthr =I0(c+t), with minimal input chargeQthr, chronaxie parameter c and total stimulation durationt as described in (Geddes, 2004). • InEquation(5.9), the parameter d is set equal to 1. The parameter τact is set equal to0.111s to reflect qualitative descriptions found in (Schmidt et al., 1996). Note: this parameter is not obtained by fitting to experimental data. • InEquation(5.10), we use a slopeλ=19.2·107 andoffset A50 =1.06·10− 6 for the brightness curve, based on a fit of our model on data by (Fernández et al., 2021). • The descriptive parameters of the distributionEquation(5.11), are set toTh50 = 9.14·10−8 andσ=6.72·10−8, based on a fit on psychometric data by (Fernández et al., 2021). • InEquations(5.12) and (5.13), we useτtrace =1.97∗10 3s andκ=14.0. These values are based on a fit of our model to data from (Schmidt et al., 1996). 5.3. Results In this section we present the results of computational experiments and comparisons with the literature to validate the biological realism, the performance, and the practical usability of the simulator. 5.3.1. Biological plausibility Here, we report on experimental data obtained from the literature, and evaluate the capacity of our simulator of fitting these empirical data. UsingEquations (5.7) to (5.10), the simulator accurately reproduces the relative phosphene brightness that was reported in the previous study for different stimulation amplitudes (R2 =0.950; verification on the same data that was used to fit the model ; Figure5.2). Figure5.3visualizes the effect of changing the stimulation parameters on the probability of phosphene perception, as estimated by our model. We compare our estimates with data reported by (Fernández et al.,2021). The model accurately fits the reported effect of pulse width, frequency and train duration on the probability of phosphenes perception. To evaluate the robustness of this fit, a 3-fold cross-validation was performed (Figure5.3panel (d)) where part of the data were held-out (i.e., the data in this panel were predicted and not in the data set used for the fit). In this more strict analysis, the prediction performance is still accurate (average R2 =0.844). FigureS4(not part of the main analysis) displays the cross-validation results after fitting the simulator to thresholding data from clinical studies that used cortical surface electrodes (ECoG) (Dobelle & Mladejovsky, 1974; Girvin et al., 1979; Niketeghad et al., 2020).
RkJQdWJsaXNoZXIy MTk4NDMw