This systematic review was conducted following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines (Page et al., 2021). Relevant studies were identified using the online search databases EMBASE, MEDLINE and PsychINFO. The following keywords were used: {(exp *blindness/) AND [(exp dream/OR exp dreaming/) OR (exp REM sleep/) OR (*vision/OR exp visual hallucination/OR exp visual illusion/)]} in EMBASE, {(exp *Blindness/) AND [(exp Dreams/OR exp Sleep, REM/) OR [(Visual Hallucination* OR Visual Perception).mp]]} in MEDLINE, and {(exp *blind/) AND [(exp dreaming/OR exp dream analysis/) OR (exp rem sleep/OR exp rapid eye movement/OR exp rem dreams/) OR (exp visual hallucinations/OR *visual perception/)]} in PsychINFO. As each database employed different subject terms, different search string was chosen for each of the three databases (please see Table 1). Eligible studies were extracted from January 1960 to March 2021. The references of the selected studies were also examined to retrieve documents missed by the literature search.

All databases were screened following the same protocol. The studies were included if they came from (1) original research articles published by a peer-reviewed scientific journal; (2) were written in English and (3) if they included observational, descriptive, longitudinal, retrospective, cross-sectional, or cohort studies. Included studies were required to investigate presence of visual imagery in congenitally blind participants, or early blind individuals with blindness of peripheral origin. Studies were excluded if they were case reports, or if they included only late blind participants. Two reviewers (NL and RB) independently screened each eligible study, and disagreements were resolved through discussion after retrieving full text to determine whether inclusion and exclusion criteria were met or consulting a third independent investigator (KI) (please also refer to Table 2). The PICOS statement is available in Table 3 and PRISMA 2020 flow diagram of study selection process is shown in Figure 1.

The following data were extracted: type of study, sample size, type of blindness, sex, age and handedness of participants; techniques and measures employed; main findings and study limitations (also please refer to Supplementary Tables 1–5). Subsequently, the studies were classified into five predetermined categories, based on the investigating techniques employed in the studies (i.e., neurotechnology, psychological investigations, sensory substitution or near-death and out-of-body experiences). A third independent investigator went through all extracted data and tables to reconcile any eventual disagreements.

The changes in human brain in subjects that are either blind from birth or very early age (congenitally blind and early blind) and in those with blindness onset later in life (late blind) were investigated by several studies. Number of neuroimaging studies in blind report different patterns of activation in primary visual and associative cortexes, as well as in prefrontal cortex (Supplementary Table 1). Röder et al. (1997) used electroencephalography in congenitally blind and sighted control subjects and detected a marked slow negative potential over the occipital cortex of 15 congenitally blind subjects who were performing mental rotation and comparison of images explored haptically. This slow negative potential was not observed in sighted control subjects who performed the same tasks. In two similar studies by De Volder et al. (2001) and Vanlierde et al. (2003), positron emission tomography was used in early blind and sighted control subjects who were performing a visuo-spatial imagery task. In both subject groups, De Volder et al. (2001) observed activation in lateral occipito-temporal areas; Vanlierde et al. (2003) in the precuneus, superior parietal lobe and occipital gyrus. Early blind group demonstrated stronger occipital activations (Vanlierde et al., 2003), whilst activation of the right prefrontal cortex and left posterior cingulate gyrus was observed only in the sighted control group (De Volder et al., 2001).

Only two studies used functional magnetic resonance imaging to explore brain activity in blind in response to specific concepts (Striem-Amit et al., 2018; Mattioni et al., 2020). Mattioni et al. (2020) found that the ventral occipito-temporal cortex encodes different concepts in a similar way in early blind (to whom concepts were presented acoustically) and in sighted control subjects (to whom concepts were presented visually). However, in the other functional magnetic resonance imaging study (Striem-Amit et al., 2018), a difference was spotted between congenitally blind and sighted controls: while in both groups, abstract concepts (e.g., “freedom”) were correlated with activity in the lateral anterior temporal lobe, and concrete concepts (e.g., “cup”) with activity in medial anterior temporal lobe, concepts such as “red,” which are imperceptible for blind people, were correlated with activity in left dorsal anterior temporal lobe in the congenitally blind group only. In the same study by Striem-Amit et al. (2018), resting-state functional connectivity analysis revealed that, in both congenitally blind and sighted control, dorsal and lateral anterior temporal lobe are connected with areas involved in semantic, non-sensorially derived information (e.g., the inferior frontal lobe), while the medial anterior temporal lobe is connected to multisensory object-related regions in central visual cortex, and in frontal and parietal lobes.

In the Supplementary Table 2, 11 studies are reported, in which mental representations of patterns, pathways, objects and spatial relations were compared between groups of congenitally blind, early blind, late blind and sighted control subjects. In three of those studies, blind subjects performed better or as well as sighted controls in visuo-spatial imagery tasks. For instance, after administering the Onomatopoeia and Images (Torrance et al., 1973), Johnson (1980) found that congenitally blind subjects produced more visual images than the sighted controls. Hollins (1985) reports that congenitally blind and sighted control subjects were equally good at identifying objects represented as patterns of squares on 2D matrices or of cubes on 3D matrices. Tinti et al. (2006) reported that both congenitally blind and late blind subjects were better than sighted controls at drawing two pathways they had just learned to walk through; however, when drawings of the two pathways were analyzed together, only late blind performed better than sighted controls.

In five of the eight remaining studies, blind and sighted subjects performed similarly well, although there were some interesting differences. For example, in one study (Kerr, 1983), the congenitally blind performed similarly to the sighted at imagining objects and their properties, although they struggled when they had to imagine concealed objects. Heller et al. (1996) found evidence of foreshortening in sighted control subjects only; however, drawings of a panel at different angular orientations were not significantly different between the early blind, the late blind and the sighted controls. In Vanlierde and Wanet-Defalque (2004), early blind, late blind and sighted control individuals performed similarly in a series of pattern recognition tasks, but they used different strategies: all 27 sighted controls, nine out of ten late blind and one out of ten early blind subjects used a “visual” strategy, while the remaining nine early blind subjects used a “coordinate” strategy, and the last late blind utilized a “mixed” (visual and coordinate) strategy. Noordzij et al. (2007) report that only their late blind–but not the early blind–group performed worse than sighted controls in a visual imagery task, while the early blind performed worse than the sighted in a spatial imagery task. In Cattaneo et al. (2010), both early blind and sighted controls participants remembered more symmetrical than non-symmetrical patterns; however, only in the sighted controls group vertical symmetries were recalled more easily than horizontal ones.

Of the 11 sensory substitution studies selected for this review (Supplementary Table 3), five employed visual-to-tactile sensory substitution devices. Specifically, Miletic et al. (1988) used a modified version of the Optacon (Linvill et al., 1966), Sampaio et al. (2001) and Chebat et al. (2007) used the TDU (Bach-y-Rita et al., 1998), Velázquez and Bazán (2010) the “on-shoe tactile display” (Velázquez et al., 2009), while Nau et al. (2015) used the BrainPort Artificial Vision Device (Wicab Inc., Madison, WI, USA). In the remaining 6 studies, the authors employed a visual-to-auditory sensory substitution devices. Renier et al. (2005), Renier et al. (2006), and Renier and De Volder (2010) used the PSVA (Capelle et al., 1998), while Abboud et al. (2014), Buchs et al. (2015), and Buchs et al. (2019) employed EyeMusic, an sensory substitution devices which uses different musical instruments to convey color information (Abboud et al., 2014).

Ring and Cooper (1997) interviewed 14 early blind, 11 late blind and 6 visually impaired subjects who had at least one near-death experience or one out-of-body experience, and found that 25 (80%) of their 31 subjects–9 (64%) out of 14 early blind subjects–claimed sight during near-death experiences and/or out-of-body experiences. These visual experiences included seeing one’s own physical body, going through a tunnel or dark space, and seeing a radiant light. Interestingly, when asked to compare near-death experiences and out-of-body experiences with their usual dream content, early blind individuals answered that visual experiences were present only in their near-death experiences/out-of-body experiences.

Several studies have demonstrated that congenitally blind people can experience oneiric visuo-spatial imagery in a way similar to sighted individuals (see Figure 2 and Supplementary Table 5; Bértolo et al., 2003, 2017; Bértolo, 2005). Moreover, some congenitally blind subjects have also been able to represent the visual content of their dreams in drawings which were similarly accurate, if somewhat less detailed, and slightly more symbolic and archetypal, as those of sighted controls (Bértolo et al., 2003). Also, in keeping with positive findings by Bértolo et al. (2003, 2017) and Bértolo (2005), some of the past studies concomitantly report findings at odds with their own negative conclusion that congenitally, and early blind subjects completely lack oneiric visual imagery. For instance, in Hurovitz et al. (1999), one of the early blind participants, totally blind since the age of 4, reported “not metaphorical” visual imagery in his dream reports; in Christensen et al. (2019), visual dream elements were reported by congenitally blind subjects on two occasions; Meaidi et al. (2014) write that “many of the blind participants in our study described an object or a scene verbally in such rich visual terms that the interlocutor began to doubt if these individuals really lacked vision.”

Whilst numerous studies in animals have elucidated how the visual cortex is developed, and how it changes in cases of monocular or binocular deprivation during different periods after birth, relative sparsity of human studies and technologic limitations have so far prevented us from gaining similar insight into human development. A complex picture emerges from reported studies: during visuo-spatial imagery tasks, the occipital cortex is activated in the blind more than in the sighted (Röder et al., 1997; Arno et al., 2001; Vanlierde et al., 2003; Collignon et al., 2007; Striem-Amit et al., 2012; Striem-Amit and Amedi, 2014), and possibly in substitution for prefrontal activity, which was sometimes detected in the sighted but not in the blind (Arno et al., 2001). However, in the occipital, temporal and parietal lobes of blind subjects, higher-order areas, which are involved in multisensory integration and supramodal representations, seem to generally preserve their function and connectivity (Arno et al., 2001; Vanlierde et al., 2003; Amedi et al., 2007; Striem-Amit and Amedi, 2014; Striem-Amit et al., 2018; Mattioni et al., 2020). Nonetheless, several distinctive features have been recorded in the blind; for instance, representations of concrete concepts that are perceptible by the sighted but not by the blind have been shown to involve activations in the left dorsal anterior temporal lobe in blind subjects only (Striem-Amit et al., 2018).

In blind subjects, the occipital cortex is activated during haptic (Amedi et al., 2010) and auditory (Kujala et al., 1995, 2005; Liotti et al., 1998; Weeks et al., 2000) processing, including pitch memory tasks (Gaab et al., 2006); “visual” cortex activation in the blind has been associated with episodic retrieval (Raz et al., 2005) and with Braille reading (Sadato et al., 1996); transcranial magnetic stimulation (TMS) used to disrupt occipital activity leads to errors during Braille reading in blind subjects, but not in sighted people (Cohen et al., 1997). Moreover, when used to stimulate occipital areas, TMS induces subjective sensations (qualia) in the fingertips of blind subjects, and then in the tongue after training with the sensory substitution device TDU (Kupers et al., 2006).

This is of note, as the geniculo-striate system is atrophic in the blind, with up to 25% volume decrease in the BA17 cortical region (Ptito et al., 2008). The creation of new corticocortical and/or thalamocortical connections, or the unmasking of pre-existing connections, normally inhibited in the presence of vision, has been suggested as a potential mechanism behind the involvement of the “blind” visual cortex in non-visual sensory processing (Sadato et al., 1996; Cohen et al., 1997; Liotti et al., 1998; Arno et al., 2001; Burton, 2003; Kupers et al., 2006, 2011; Müller et al., 2019). Indeed, in animals, such as for example blind moles, the inferior colliculus projects not only to the auditory medial geniculate nucleus, but also to the visual lateral geniculate nucleus of the thalamus (Doron and Wollberg, 1994). In humans, diffusion tensor imaging revealed microstructural changes within thalamic clusters in blind individuals, even when no gross differences in the thalamocortical network were observed (Reislev et al., 2017). In keeping, Müller et al. (2019) have proposed the existence of a thalamocortical pathway for fast rerouting of haptic information to the blind “visual” cortex, activated only 35 ms after haptic stimulation in congenitally blind subjects, but not in blindfolded sighted people.

Different strategies in perceiving objects by sighted and blind were previously argued by Révész and Wolff (1950) in Psychology and Art of the Blind. He distinguished between “Haptics” and “Optics,” which, in his opinion, lead to different representations of objects, shapes and space in blind and sighted subjects. He argued that an object can be immediately perceived in its entirety through vision, while it can only be gradually “discovered” in a series of subsequent haptic explorations by a blind person. Thus, according to Révész and Wolff (1950), the same object must be represented in qualitatively different ways between the blind and sighted individuals. Indeed, considering diverging findings from analyzed studies, it remains unclear which mechanistic processes may support the presence and nature of visuo-spatial imagery in blind individuals. It is obvious that congenitally blind and early blind subjects may have some mental representation of patterns, shapes and spatial relations. Nonetheless, it is not clear how much these representations are affected by the blind’s main perceptual modality (i.e., haptics), and therefore how much visuo-spatial representations in the blind differ from those of the sighted. For instance, it appears that while congenitally blind and early blind subjects are as good as–or even better than–sighted controls at drawing pathways, imagining objects and remembering 2D patterns, they struggle to imagine concealed objects and reproduce 3D patterns, and they lack foreshortening.

A century ago, Romains (1920) managed to convey visuo-spatial information through the skin of blindfolded sighted subjects. Later in the 20th century, Starkiewicz and Kuliszewski (1963) developed the “Elektroftalm,” a device which transformed visuo-spatial images into patterns of vibrations, enabling blind subjects to “see.” Another pioneering method of sensory substitution was trialed by Bach-y-Rita et al. (1969), whose research group built an sensory substitution device consisting of a camera mounted on a dental chair. Here, visual information from the camera was converted to vibratory stimulation of the back of blind people sitting on the chair (Bach-y-Rita et al., 1969). Since then, the field of sensory substitution has grown and seen the development of various sensory substitution devices, not only for the blind; see for example Riso (1999) or Tyler et al. (2003).

In this review, we included 11 recent studies in which visual-to-auditory or visual-to-vibrotactile sensory substitution devices were employed to investigate visuo-spatial perception and imagery in congenitally and/or early blind subjects. Overall, it would appear that congenitally blind, early blind and late blind, as well as sighted subjects, can integrate non-visual information (auditory and haptic in this case) to create visuo-spatial mental images. However, there are differences between vision as a result of visual inputs and “blind vision” acquired through sensory substitution. For example, blind subjects using sensory substitution devices were shown insensitive to either Ponzo (Renier et al., 2005) or vertical-horizontal (Renier et al., 2006) illusions, and were similarly shown to have difficulties recognizing 3D shapes and spatial relations (Miletic et al., 1988; Velázquez and Bazán, 2010). Thus, one could argue that whilst visual imagery is possible even in the absence of inputs from the retina, it is still influenced by the specific sensory modality in which information is conveyed to visual association areas.

Anecdotally, visuo-spatial imagery has been reported during near-death and out-of-body experiences by the blind, albeit the mechanistic nature of any such imagery remains poorly understood (Ring and Cooper, 1997). Near-death or out-of-body experience are rare parapsychological phenomena reported to occur during clinical cardio-respiratory arrest, or other drug or accident induced states, when a person seems to be awake and sees his body and the world from a location outside his physical body (Bünning and Blanke, 2005) [for an in-depth overview of the topic, also see Mindsight: Near-Death and Out-of-Body Experiences in the Blind (Ring and Cooper, 1999)].

The above body of data may be taken to support the notion that (oneiric) mental representation of images may not be dependent on visual input, and that they can therefore be a, (albeit infrequent) part of congenitally blind individuals’ dreams.

Past evidence suggests that congenitally blind people may experience more circadian dysregulation (Lockley et al., 2007) and sleep-related problems (64% of congenitally blind subjects) than late blind or sighted controls (43 and 30%, respectively), as assessed using the Pittsburgh Sleep Quality Index. They also report suffering with more nightmares [mean frequency of nightmares over total number of dream reports in congenitally blind subjects: (25%) than late blind (7%) or sighted controls (6%) (Meaidi et al., 2014)].

However, the presence of visual imagery in the oneiric mentation in congenitally blind remains widely disputed, and highly debatable, despite some anecdotal, as well as some limited scientific evidence to the contrary (Berger et al., 1962; Amadeo and Gomez, 1966; Kirtley, 1975; Kerr et al., 1982; Hurovitz et al., 1999; Holzinger, 2000; Lopes Da Silva, 2003; Staunton and O’Rourke, 2012; Meaidi et al., 2014; Christensen et al., 2019; Andrade, 2021). It is also possible, that any such reports might have been historically ignored and disregarded (Kerr et al., 1982; Hurovitz et al., 1999; Holzinger, 2000; Staunton and O’Rourke, 2012; Meaidi et al., 2014; Christensen et al., 2019). Instead, it has been widely accepted that auditory, haptic, proprioceptive, somaesthetic, olfactory and gustatory imagery, as well as that of pain and temperature may be more prevalent in dreams of congenitally blind, by comparison to sighted people (Hurovitz et al., 1999; Holzinger, 2000; Staunton and O’Rourke, 2012; Meaidi et al., 2014; Supplementary Table 5).

When analyzing dream contents, most studies failed to record reports of visual impressions in the dreams of congenitally blind individuals (Kerr et al., 1982; Hurovitz et al., 1999; Holzinger, 2000; Staunton and O’Rourke, 2012; Meaidi et al., 2014), with only visual impressions (color, light) reported in the dreams of congenitally blind subjects with residual vision (Kerr et al., 1982; Meaidi et al., 2014). In late blind participants, the average blindness duration index has been shown to negatively correlates with duration, clarity, and color content of visual impressions in dreams (Meaidi et al., 2014). Positive correlations between lack of oneiric visual imagery and (1) congenital or early blindness, (2) total blindness, and (3) a high percentage of the subject’s life spent being blind has been reported by Hurovitz et al. (1999). Interestingly, some late blind subjects reported experiencing oneiric visual imagery (including colors) of people and objects only after the onset of blindness (Kerr et al., 1982; Holzinger, 2000). In that respect, it is of note that Kirtley (1975) proposed that, if the onset of blindness is before the age of five, no visual experience in the dreams should be possible, whilst onset of blindness between the ages five and seven may produce some oneiric visual imagery. It has been thus proposed that visual imagery in dreams will be retained only if onset of blindness happens after age of seven. Conversely, in one study by Hurovitz et al. (1999) it has been suggested that oneiric visual impressions may be retained if sight was lost after the age of four. Moreover, Meaidi et al. (2014) argued that only individuals who were born blind (congenitally blind), or who lost their sight before 2.5 years of age, will completely lack visual imagery in their dreams.

In this background, it is also of note that congenitally blind people have been shown to have significantly reduced, or absent, rapid eye movements (REM) during sleep (Christensen et al., 2019), whilst they appear to retain some of the other EEG features of phasic REM sleep microstructure. The phasic periods of REM sleep are commonly discerned by bursts of eye movements linked to so-called ponto-geniculo-occipital waves, contractions of the middle ear muscles, myoclonic twitches of skeletal muscles, sawtooth waves, as well as irregularities in cardio-respiratory activity (Simor et al., 2020). Some authors have suggested that the phasic REM sleep may play its own specific role in dreaming and emotion regulation processes during sleep, likely relating to reactivation of vivid visuo-spatial (emotionally relevant) mental pictures, or to the reprocessing of emotional memories during sleep (Simor et al., 2020). Thus, arguably, in blind, differential neurocircuitry may be activated during the phasic REM to that in sighted individuals, possibly reflecting use of auditory (or haptic), instead of visual referencing points. However, it remains unclear whether this may also translate in differential functional role of this fundamental physiologic process.

This is of particular importance given that, sleep itself, and more specifically, REM sleep (Marks et al., 1995) is crucial for the neurodevelopment of the visual cortex (Frank et al., 2001; Hobson, 2009), and therefore of mental imagery (Foulkes, 1982, 1999; Hobson, 2009; Siclari et al., 2020).