5.4. Discussion 5 75 habituation. Possibly, a more complex, non-linear, model can more accurately fit the neurophysiological data, as discussed in sectionSection 5.4.3. However, the presented model has the benefit of simplicity (there are only three parameters). Also, note that there is still some ambivalence in the clinical data. In contrast to the findings by (Schmidt et al., 1996), some studies have found accommodation over different time scales and sometimes no accommodation at all (Bak et al., 1990; Bartlett et al., 1977; Dobelle & Mladejovsky, 1974; Fernández et al., 2021). More research is required for a better understanding of the neural response after repeated or prolonged stimulation. Phosphene shape and appearance The appearance of phosphenes in our simulation (white, round, soft dots of light) are largely in line with previous reports on intracortical stimulation (Bak et al., 1990; Fernández et al., 2021; Schmidt et al., 1996). However, more elongated shapes and more complex shapes have been reported as well (Bak et al., 1990; Bosking et al., 2017a; Winawer & Parvizi, 2016). By using separately generated phosphene renderings, our simulator enables easy adjustments of the appearance of individual phosphenes. Additionally, we incorporated the possibility to change the default Gaussian blob appearance into Gabor patches with a specific frequency and orientation. Regarding the colour of phosphenes, there is still some ambivalence in the reports, including descriptions of phosphene color ranging from black or white to different tones of color (Schmidt et al., 1996; Tehovnik & Slocum, 2007; Tehovnik et al., 2009). Notably, increasing the stimulation amplitudes can lead the appearance to shift from colored to yellowish or white (Schmidt et al., 1996). This effect may be explained by the increased current spread for higher stimulation amplitudes, which is predicted to span multiple cortical columns coding for different visual characteristics (e.g., orientation or colour), thus giving rise to phosphenes with amalgamated features (Tehovnik & Slocum, 2007). Currently, the limited amount of systematic data render it difficult to enable more accurate simulations of the variability in phosphene appearance. 5.4.2. End-to-end optimization Dynamic encoding The results presented inFigure5.7demonstrate that our proposed realistic phosphene simulator is well-suited for the dynamic optimization in an end-to-end architecture. Our proof-of-principle video-encoding experiments are the first to explicitly optimize the stimulation across the temporal domain. This provides a basis for the further exploration of computationally-optimized dynamic stimulation patterns. Dynamic optimization of the stimulation may be necessary to counteract unwanted effects such as response fading due to accommodation after repeated or prolonged stimulation (Schmidt et al., 1996), or delayed phosphene perception after stimulation on- and offset. The inclusion of a realistic simulator in the optimization pipeline enables researchers to exploit the optimal combination of stimulation parameters to obtain precise control over the required perception. Moreover, besides acquiring optimal control over the transfer function from stimulation to phosphenes, dynamic phosphene encoding could also prove useful to expand the encoded information along the temporal domain (Beauchamp et al., 2020). Although this was not in the scope of the current study, our software is well-suited for
RkJQdWJsaXNoZXIy MTk4NDMw