Huntington’s Disease (HD) is an autosomal dominant neurodegenerative disorder caused by a repeated expansion of the CAG trinucleotide (≥ 39 repeats) in the Huntingtin gene (Tabrizi et al., 2019). It is characterized by motor, cognitive, and psychiatric symptoms. The evolution of HD includes an asymptomatic pre-manifest phase (preHD) and a manifest phase with symptoms gradually affecting patients’ daily living, before leading to death within 35 years after clinical onset (Walker, 2007).

Among daily impairments in HD, individuals have reported spatial abilities deficits, which refers to the way individuals mentally represent and manipulate spatial information (Shah & Miyake, 2005). These deficits include visuoperceptual, visuoconstructive, visual scanning, and mental rotation difficulties, starting before clinical onset. This causes confusion in daily functioning, which can influence overall cognitive performances (Coppen et al., 2018). Although spatial disorders are less common in HD compared to diseases like Alzheimer’s, it is likely undervalued, especially since significant occipital atrophy is evident even in the preHD stage of the disease (Scahill et al., 2013). However, due to the challenges associated with monitoring spatial capacities, this atrophy and its consequences have not been thoroughly explored.

Spatial deficits are traditionally measured with pen-and-pencil tests during neuropsychological assessments (Brandt et al., 2004; Gómez-Tortosa et al., 1996; Lawrence et al., 2000). However, individuals with HD often face limited access to appropriate healthcare systems and lengthy comprehensive assessments involving a multidisciplinary clinical team. This can lead to poor symptom management and a reduction of individuals’ quality of life (Frich et al., 2016; van Lonkhuizen et al., 2023). Because it is easily collectible, brief, and examiner-free, speech is thought as the best candidate to assess patients’ performances remotely and will see greater utilization in future healthcare systems (Fagherazzi et al., 2021; Robin et al., 2020). In HD, it has been shown that speech can predict patients’ disease progression, motor, and functional performances, as well as cognitive functions such as attention, executive functions, and cognitive flexibility (Riad et al., 2020, 2022; Romana et al., 2020; Vogel et al., 2012). However, to make it a more complete tool and optimize its future use, it remains to assess its ability to test HD patients' visuospatial performance.

Several studies have yet demonstrated that spatial language is a specific domain within language. Spatial language translates spatial representation in language by using spatial terms, i.e., spatial prepositions and action verbs, to describe the spatial relations between objects (Landau & Jackendoff, 1993; Chatterjee, 2008). Descriptions of objects’ spatial relations are influenced by the referential position of the object in space, with entities positioned on the left being more frequently mentioned first than those on the right in a scene in healthy individuals (Baltaretu et al., 2016). Processing spatial language involves brain activity in the occipito-parietal stream which supports the “where pathway”, whereas object semantics activates the occipito-temporal stream which supports the “what pathway” in neuroimaging studies (Conder et al., 2017; Rocca et al., 2020). The ability to use appropriate language to describe a visual scene correlates with the acquisition of spatial abilities in children (Bowerman et al., 1995; Miller et al., 2017), as well as its decline with normal aging showing that language grasps visuo-spatial capacities (Ardila & Rosselli, 1996; Markostamou & Coventry, 2022). Likewise, cerebral lesions may yield difficulties in production and understanding spatial terms (Tranel & Kemmerer, 2004). Lastly, language allows the encoding of spatial representation, as demonstrated by the ability to create geographical maps from written corpora (Louwerse & Zwaan, 2009; Friedman et al., 2002). Embodiment based on sensorimotor interaction could also offer a conceptual spatial shortcut in the hypothesis of symbolic interdependence (Louwerse, 2011). Altogether, these findings suggest that spatial language reflects individuals’ spatial abilities.

This has triggered the use of the Cookie Theft Picture (CTP) description task to assess spatial abilities through, in which participants are required to scan, infer relationships between pictorial elements, and report them orally (Goodglass & Kaplan, 1972). Thus, three studies were run in Alzheimer's disease, each focusing on various aspects of spatial language (Figure 1). First, the Spatial neglect method enabled classifying Alzheimer’s individuals from controls by counting the number of times each pictorial elements are mentioned in the different divisions of the picture (Masrani, 2018). Attention, concentration, repetition, and perception scores were then calculated based on the occurrence of the mentioned pictorial elements without considering the spatial relations between the elements in the descriptions. Secondly, Bosse’s study described the frequency of spatial prepositions that characterized location and direction of pictorial elements in Alzheimer’s individuals CTP descriptions. The frequency was lower for left/right, directional, and dynamic prepositions compared to healthy individuals. However, action verbs were not assessed, excluding some spatial and directional information held by these action verbs (Bosse, 2019). Because both methods, partially integrated spatial context, Ambadi et al. proposed the Spatio-semantic model using graph theory for Alzheimer’s CTP description analyses. Graph theory is a mathematical framework, which represents a network of nodes (the pictorial elements like objects, people, location etc.) connected by arrows according to the order of the description (called edges in graph theory). Whereas this model distinguishes Alzheimer’s and healthy individuals by reflecting their descriptions’ visual narrative paths, attentional, and organizational abilities (Ambadi et al., 2021), it does not assess spatial prepositions and action verbs. Therefore, in “The boy is wearing trousers and the mother a dress. This girl is on the left side of the boy, wearing a skirt”, the Spatio-semantic approach models the entire path between the mentioned elements encompassing the single spatial relation with the non-spatial descriptions (Boy → Mother → Girl → Boy), thus accounting for three semantic relations rather than a single spatial one. This might artificially increase the representation of each element as a spatial object despite the fact the utterance is not expressing a spatial relation and could constitute a caveat in perseverative patients.

Therefore, considering the low spatial disturbances and high perseverative behavior in HD compared to AD (Coppen et al., 2018; Rosenblatt et al., 2007), to ensure capturing the spatial deficit in HD strengthening the Spatio-semantic model appears necessary. Here, we thus developed the Spatial Description Model, a comprehensive language-based assessment of spatial disturbances, selectively considering only sentences that contain spatially relevant information before modeling them graphically. The Model operated by removing sentences without any information about the pictorial elements’ location or their spatial relations. It maximized the extraction of spatial information by integrating together the mentioned pictorial elements with spatial language. Spatial relations were identified by detecting labeled spatial terms, pictorial elements, and determining if they were used together to describe the picture spatially. Thus, using the previous example, our algorithm would identify only a single spatial relation (Girl → Boy) rather than three. To validate our Model, we assessed its convergence/divergence validity using the SelfCog battery in a cross-sectional cohort of HD carriers and control participants (Lunven et al., 2023). In addition, we checked the added value of our model by comparing its performance with the ones developed in Alzheimer’s disease (Figure 1).