Patient conditions vary greatly, requiring the clinician to use their expertise to manage these differences. They must determine whether variations signify different diseases requiring unique treatments or are just expected differences within the same disorder requiring similar treatment approaches. There are three types of variations: categorical, dimensional, and random noise. Categorical variation refers to clear-cut patient differences, making it easy to understand and research. This is what traditional diagnosis methods are based on. Dimensional variations involve measurable factors such as the severity of a disorder or common dimensions that cut across different conditions, like blood pressure and body temperature. Recent studies have highlighted the benefits of multidimensional behavioural geometries over categorising every variation in a patient's behaviour. These methods allow for a more nuanced understanding by mapping patients' symptoms onto multidimensional scales, capturing broad distinctions, subtle variations, mixed features, and changes over time. Multidimensional approaches help identify common symptoms across different disorders that are important for predicting outcomes and could potentially be treated, like apathy in neurodegenerative diseases.
This study focused on the challenge of how to relate behavioural dimensions to underlying neurological changes in patients. The study applied multidimensional behavioural geometries to represent the variations in patients' symptoms and relate them to atrophy patterns. Frontotemporal dementia (FTD) cortical and subcortical syndromes, characterised by progressive cognitive, behavioural, language, and motor deficits, were used as test cases. The investigation achieved this aim by 1) Using principal component analysis (PCA) to extract the multidimensional behavioural geometry from the Mini-Language State Examination (MLSE) battery, thereby capturing the continuum-graded language variations across the FTD spectrum. 2) Employing three approaches to reveal the brain-behaviour relationships: (i) classic univariate voxel-based morphometry correlations; (ii) PCA of atrophy maps to extract coherent brain patterns and thus test which of these disease-defined anatomical dimensions relate to the behavioural dimensions; and (iii) exploring whether variation on the behavioural dimensions relates to the integrity of one versus several brain regions, and thus shifting from univariate to multivariate regression models to capture the latter cognition-to-distributed network relationships, when they exist.
Our findings revealed more distributed atrophy patterns for motor-speech/phonology and syntax factors, while semantics is more localised in the anterior part of the inferior temporal lobe. Importantly, while many studies rely on univariate correlations to investigate brain correlations in cognitive disorders, our findings indicate that this approach falls short of capturing the complex relationships between behaviour and the brain. Given the distributed and interactive nature of language functions, isolating specific language processes in the brain poses a great challenge, affecting how language is altered, reflected in the heterogeneity of language deficits in FTD. Incorporating multivariate analysis becomes essential to capture these intricate associations and identify common and distinct neural mechanisms of language. In conclusion, other aspects of language besides semantics rely on various brain regions, demonstrating the complex multidimensional architecture of language functions.
