The primary human visual pathway conveys the majority of retinal fibers to the lateral geniculate nucleus of the thalamus and then, via the optic radiations, to the primary visual cortex (V1) (the retino-geniculo-striate pathway). V1 is the main distributor of visual information to extrastriate visual areas, for further processing. A secondary visual pathway (the retino-collicular pathway) routes a minority of retinal fibers directly to the Superior Colliculus (a midbrain structure), which also has reciprocal connections with striate and extrastriate visual cortices (May, 2006).
Patients with lateralized damages to the primary visual cortex (V1) or to the neural pathway feeding V1 often develop homonymous hemianopia, a visual field defect with the loss of conscious vision in one hemifield. Hemianopic patients cannot perceive visual stimuli presented in the blind hemifield; moreover, they show the inability to spontaneously develop effective oculomotor strategies to compensate for the visual field loss (Hildebrandt et al., 1999; Zihl, 2000; Tant et al., 2002).
Despite the visual deficit, hemianopic patients can preserve the ability to integrate audiovisual stimuli in the affected field, with beneficial effects (Frassinetti et al., 2005; Leo et al., 2008). In particular, data by Frassinetti and colleagues (Frassinetti et al., 2005) show that patients performing a visual detection task, while maintaining central fixation, significantly improved conscious visual detections in the affected field, when the auditory stimuli were applied in spatial and temporal coincidence with the visual targets.
The Superior Colliculus is the most likely structure mediating this multisensory improvement, because of its anatomical connections and the properties of its neuronal responses. Indeed, SC neurons receive not only visual information but also signals from other different sensory modalities, such as audition (Meredith and Stein, 1986; Stein and Meredith, 1993; May, 2006). Visual and auditory information are integrated in multisensory SC neurons according to specific principles (Stein and Meredith, 1993): an audiovisual stimulation elicits a stronger neuronal activation than each single component, when the visual and auditory components are presented in spatial and temporal register (spatial and temporal principle). Moreover, a proportionally greater enhancement of multisensory neuronal activation is evoked when weakly effective unisensory stimuli are combined, compared to the combination of highly effective stimuli (inverse effectiveness principle). The SC integrative principles have strong implications in hemianopia, as the SC and the retino-collicular pathway are preserved in these patients. Visual retinal input to SC, although weak, can still be efficiently combined with an accessory auditory input thanks to the inverse effectiveness principle, provided the rule of spatial and temporal proximity is satisfied. Furthermore, SC multisensory enhancement can affect cortical visual processing thanks to the projections from the SC to the visual cortices.
In addition to the immediate, “online” multisensory improvement in visual detection, there is also evidence of prolonged, “offline” effects that can be induced by repeated exposure to audiovisual stimuli. Indeed, long-lasting improvements of visual performances in hemianopic patients, promoted by audiovisual training protocols stimulating the blind hemifield, have been reported (Bolognini et al., 2005; Passamonti et al., 2009; Dundon et al., 2015b; Tinelli et al., 2015; Grasso et al., 2016). During the training, a visual target was given in close spatial and temporal proximity with an auditory stimulus, at various positions in the visual field; patients were asked to detect the presence of the visual target, by directing the gaze toward it from a central fixation point. Results revealed a significant post-training improvement in detection of unimodal visual targets in the blind field when the patients were allowed to use eye movements, while a weak amelioration was found when they had to maintain central fixation (Bolognini et al., 2005; Tinelli et al., 2015). Such results suggest that the audiovisual training could promote an increased oculomotor response to visual stimuli in the affected hemifield.
Beyond the “online” effects of audiovisual stimulation, the Superior Colliculus is a possible candidate for mediating the training effects, too. Indeed, the SC projects to brainstem motor areas controlling eyes and head orientation, and is critically involved in the initiation and execution of reflexive (i.e., exogenously-driven) saccades (Sparks, 1986; Jay and Sparks, 1987a; May, 2006; Johnston and Everling, 2008). Importantly, more than 70% of SC neurons projecting to the brainstem and, therefore involved in saccade generation, respond to multisensory stimulations (Meredith and Stein, 1986). As such, audiovisual stimuli, enhancing multisensory SC activation, might plastically reinforce the gain of the transduction from the SC sensory response to the motor output; in other words, after training the oculomotor system could have acquired increased responsiveness to the visual input conveyed via the retino-collicular pathway. However, the plastic mechanisms and synaptic reorganization that can functionally instantiate these visuomotor capabilities remain undetermined. Moreover, it is unclear whether the training may even stimulate genuine visual restitution beyond oculomotor compensation, and how the compensatory and restitutive effects may complementary contribute to visual improvements.
Recently, we have developed a neural network model (Magosso et al., 2016) that formalized the main cortico-collicular loops involved in audiovisual integration, and implemented—via neural connections and input-output neural characteristics—the SC multisensory integrative principles. The network postulated neural correlates of visual consciousness and mimicked unilateral V1 lesion. Simulations, performed in fixed-eyes condition, reproduced the “online” effects of enhanced visual detection under audiovisual stimulation.
Here, we extend our previous neural network to explore the effects of training in simulated hemianopic patients, providing quantitative predictions that can contribute to a mechanistic understanding of visual performance improvement observed in real patients. To this aim, the network has been integrated by novel elements. First, we have included a module of saccade generation, embracing the colliculus sensory-motor transduction; in this way, we can account for the potentiality of short-latency saccades triggered in a bottom-up fashion. Second, Hebbian mechanisms of synaptic learning have been implemented and adopted during training simulations. Different training paradigms (audiovisual multisensory/visual unisensory stimulation in eye-movements/fixed-eyes condition) are tested, to examine their efficacy in promoting different forms of rehabilitation (compensatory/restitutive), and to assess the predicted results in light of in vivo data.