Patterns of brain volume and metabolism predict clinical features in the progressive supranuclear palsy spectrum

Abstract

Progressive supranuclear palsy (PSP) is a neurodegenerative tauopathy that presents with highly heterogenous clinical syndromes. We perform cross-sectional data-driven discovery of independent patterns of brain atrophy and hypometabolism across the entire PSP spectrum. We then use these patterns to predict specific clinical features and to assess their relationship to phenotypic heterogeneity. We included 111 patients with PSP (60 with Richardson syndrome and 51 with cortical and subcortical variant subtypes). Ninety-one were used as the training set and 20 as a test set. The presence and severity of granular clinical variables such as postural instability, parkinsonism, apraxia and supranuclear gaze palsy were noted. Domains of akinesia, ocular motor impairment, postural instability and cognitive dysfunction as defined by the Movement Disorders Society criteria for PSP were also recorded. Non-negative matrix factorization was used on cross-sectional MRI and fluorodeoxyglucose-positron emission tomography (FDG-PET) scans. Independent models for each as well as a combined model for MRI and FDG-PET were developed and used to predict the granular clinical variables. Both MRI and FDG-PET were better at predicting presence of a symptom than severity, suggesting identification of disease state may be more robust than disease stage. FDG-PET predicted predominantly cortical abnormalities better than MRI such as ideomotor apraxia, apraxia of speech and frontal dysexecutive syndrome. MRI demonstrated prediction of cortical and more so sub-cortical abnormalities, such as parkinsonism. Distinct neuroanatomical foci were predictive in MRI- and FDG-PET-based models. For example, vertical gaze palsy was predicted by midbrain atrophy on MRI, but frontal eye field hypometabolism on FDG-PET. Findings also differed by scale or instrument used. For example, prediction of ocular motor abnormalities using the PSP Saccadic Impairment Scale was stronger than with the Movement Disorders Society Diagnostic criteria for PSP oculomotor impairment designation. Combination of MRI and FDG-PET demonstrated enhanced detection of parkinsonism and frontal syndrome presence and apraxia, cognitive impairment and bradykinesia severity. Both MRI and FDG-PET patterns were able to predict some measures in the test set; however, prediction of global cognition measured by Montreal Cognitive Assessment was the strongest. MRI predictions generalized more robustly to the test set. PSP leads to neurodegeneration in motor, cognitive and ocular motor networks at cortical and subcortical foci, leading to diverse yet overlapping clinical syndromes. To advance understanding of phenotypic heterogeneity in PSP, it is essential to consider data-driven approaches to clinical neuroimaging analyses.

Keywords: diagnosis; machine learning; neurodegeneration; neuroimaging; tauopathy.

Graphical Abstract

Figure 1

Overview of our approach with concrete examples of methods. The top panel shows a process map outlining our methods, with numbers corresponding to the detailed description in this caption. For some numbered items, we have provided concrete examples in the panels below. Starting with the entire cohort, we split off 20 participants with both MRI and FDG available to serve as a test set. We then fit the non-NMF model in the training set, using cross-validation reconstruction error to determine the NMF rank (1). The optimal NMF model is then fit to the training data, resulting in a set of components and participant loads on these components (2). An example of one MRI component is shown (2), with high weights in the midbrain. An example of a participant with a low load and a high load on this component is shown, with low loads corresponding to midbrain atrophy. The NMF model is then applied to the training set (3), resulting in loads for those participants. For each clinical metric of interest, we fit elastic net regression models using loads on all components as predictors (4). To visualize the voxels that contribute the most to a prediction, we create ‘model coefficient maps’ by multiplying the slope of each component load term in the model by the corresponding component weight image and summing these (4). The fitted elastic net models are then applied to the test set (5). For both test and train, the fitted models produce predictions, and these are compared to the true values to assess performance (6). AOS, apraxia of speech; CV, cross-validation; FDG, flouro-deoxy glucose; MRI, magnetic resonance imaging; NMF, non-negative matrix factorization; VSGP, vertical supranuclear gaze palsy.

Figure 2

Voxel-level illustration of model beta weights for binary variables present. Estimates for the MRI models are shown on the left and FDG-based models on the right. Blue voxels represent those where increased atrophy/hypometabolism was associated with a higher probability of the binary clinical feature being present. Frontal cortical involvement was associated with AOS and IMA, with FDG models showing more regional specificity—ventrolateral precentral for AOS and dorsolateral for IMA. Subcortical involvement was associated with a lower probability of AOS, possibly reflecting cortical–subcortical phenotypic variability as discussed in the text. Frontal cortical and subcortical, as well as brainstem, atrophy was associated with parkinsonism and VSGP for MRI models. For FDG models, the pattern for parkinsonism was less clear—anterior frontal and superior parietal involvement was associated with a lower probability of parkinsonism, whereas precentral and superior frontal areas were associated with increased probability. The FDG association for presence of VSGP was similar to that of MRI, but with more cortical and less subcortical weighting. AOS, apraxia of speech; FDG, flouro-deoxy glucose; MCC, Matthews correlation coefficient; IMA, ideomotor apraxia; P, parkinsonism; VSGP, vertical supranuclear gaze palsy.

Figure 3

Voxel-level illustration of model beta weights for select continuous variables. Estimates for the MRI models are shown on the left and FDG-based models on the right. Blue voxels represent those where increased atrophy/hypometabolism was associated with a higher predicted severity on the clinical feature, whereas red voxels indicate those where atrophy/hypometabolism was associated with a lower predicted severity. MRI- and FDG-based models both associated motor and premotor atrophy/hypometabolism with increased severity, but MRI additionally had widespread inverse associations, including subcortical and cerebellar atrophy being associated with less severe ASRS. This may reflect a cortical versus subcortical phenotype effect. For TULIA and MoCA, in contrast, the regions associated with increased severity were similar. Interestingly, although both MRI and FDG associated frontal hypometabolism/atrophy with worse performance on the FAB, the MRI models placed more emphasis on cerebellar voxels and the FDG models more on parietal voxels. ASRS, Apraxia of Speech Rating Scale; FAB, frontal assessment battery; FDG, fluorodeoxyglucose; MoCA, Montreal Cognitive Assessment; MRI, magnetic resonance imaging; TULIA, apraxia scale.