Generalizing post-stroke prognoses from research data to clinical data

Stroke lesion size - Still a useful biomarker for stroke severity and outcome in times of high-dimensional models

BACKGROUND

The volumetric size of a brain lesion is a frequently used stroke biomarker. It stands out among most imaging biomarkers for being a one-dimensional variable that is applicable in simple statistical models. In times of machine learning algorithms, the question arises of whether such a simple variable is still useful, or whether high-dimensional models on spatial lesion information are superior.

METHODS

We included 753 first-ever anterior circulation ischemic stroke patients (age 68.4±15.2 years; NIHSS at 24 h 4.4±5.1; modified Rankin Scale (mRS) at 3-months median[IQR] 1[0.75;3]) and traced lesions on diffusion-weighted MRI. In an out-of-sample model validation scheme, we predicted stroke severity as measured by NIHSS 24 h and functional stroke outcome as measured by mRS at 3 months either from spatial lesion features or lesion size.

RESULTS

For stroke severity, the best regression model based on lesion size performed significantly above chance (p < 0.0001) with R2 = 0.322, but models with spatial lesion features performed significantly better with R2 = 0.363 (t(752) = 2.889; p = 0.004). For stroke outcome, the best classification model based on lesion size again performed significantly above chance (p < 0.0001) with an accuracy of 62.8%, which was not different from the best model with spatial lesion features (62.6%, p = 0.80). With smaller training data sets of only 150 or 50 patients, the performance of high-dimensional models with spatial lesion features decreased up to the point of being equivalent or even inferior to models trained on lesion size. The combination of lesion size and spatial lesion features in one model did not improve predictions.

CONCLUSIONS

Lesion size is a decent biomarker for stroke outcome and severity that is slightly inferior to spatial lesion features but is particularly suited in studies with small samples. When low-dimensional models are desired, lesion size provides a viable proxy biomarker for spatial lesion features, whereas high-precision prediction models in personalised prognostic medicine should operate with high-dimensional spatial imaging features in large samples.

The role of the hippocampus in statistical learning and language recovery in persons with post stroke aphasia

Although several studies have aimed for accurate predictions of language recovery in post stroke aphasia, individual language outcomes remain hard to predict. Large-scale prediction models are built using data from patients mainly in the chronic phase after stroke, although it is clinically more relevant to consider data from the acute phase. Previous research has mainly focused on deficits, i.e., behavioral deficits or specific brain damage, rather than compensatory mechanisms, i.e., intact cognitive skills or undamaged brain regions. One such unexplored brain region that might support language (re)learning in aphasia is the hippocampus, a region that has commonly been associated with an individual's learning potential, including statistical learning. This refers to a set of mechanisms upon which we rely heavily in daily life to learn a range of regularities across cognitive domains. Against this background, thirty-three patients with aphasia (22 males and 11 females, M = 69.76 years, SD = 10.57 years) were followed for 1 year in the acute (1-2 weeks), subacute (3-6 months) and chronic phase (9-12 months) post stroke. We evaluated the unique predictive value of early structural hippocampal measures for short-term and long-term language outcomes (measured by the ANELT). In addition, we investigated whether statistical learning abilities were intact in patients with aphasia using three different tasks: an auditory-linguistic and visual task based on the computation of transitional probabilities and a visuomotor serial reaction time task. Finally, we examined the association of individuals' statistical learning potential with acute measures of hippocampal gray and white matter. Using Bayesian statistics, we found moderate evidence for the contribution of left hippocampal gray matter in the acute phase to the prediction of long-term language outcomes, over and above information on the lesion and the initial language deficit (measured by the ScreeLing). Non-linguistic statistical learning in patients with aphasia, measured in the subacute phase, was intact at the group level compared to 23 healthy older controls (8 males and 15 females, M = 74.09 years, SD = 6.76 years). Visuomotor statistical learning correlated with acute hippocampal gray and white matter. These findings reveal that particularly left hippocampal gray matter in the acute phase is a potential marker of language recovery after stroke, possibly through its statistical learning ability.

The strange role of brain lesion size in cognitive neuropsychology

The size of brain lesions is a variable that is frequently considered in cognitive neuropsychology. In particular, lesion-deficit inference studies often control for lesion size, and the association of lesion size with post-stroke cognitive deficits and its predictive value are studied. In the present article, the role of lesion size in cognitive deficits and its computational or design-wise consideration is discussed and questioned. First, I argue that the commonly discussed role or effect of lesion size in cognitive deficits eludes us. A generally valid understanding of the causal relation of lesion size, lesion location, and cognitive deficits is unachievable. Second, founded on the theory of causal inference, I argue that lesion size control is no generally appropriate covariate control. Instead, it is identified as a procedure with only situational benefits, which is supported by empirical data. This theoretical background is used to suggest possible research practices in lesion-deficit inference, post-stroke outcome prediction, and behavioural studies. Last, control for lesion size is put into a bigger historical context - it is identified to relate to a long-known association problem in neuropsychology, which was previously discussed from the perspectives of a mislocalisation in lesion-deficit mapping and the symptom complex approach.

Lesion site and therapy time predict responses to a therapy for anomia after stroke: a prognostic model development study

Stroke is a leading cause of disability, and language impairments (aphasia) after stroke are both common and particularly feared. Most stroke survivors with aphasia exhibit anomia (difficulties with naming common objects), but while many therapeutic interventions for anomia have been proposed, treatment effects are typically much larger in some patients than others. Here, we asked whether that variation might be more systematic, and even predictable, than previously thought. 18 patients, each at least 6 months after left hemisphere stroke, engaged in a computerised treatment for their anomia over a 6-week period. Using only: (a) the patients' initial accuracy when naming (to-be) trained items; (b) the hours of therapy that they devoted to the therapy; and (c) whole-brain lesion location data, derived from structural MRI; we developed Partial Least Squares regression models to predict the patients' improvements on treated items, and tested them in cross-validation. Somewhat surprisingly, the best model included only lesion location data and the hours of therapy undertaken. In cross-validation, this model significantly out-performed the null model, in which the prediction for each patient was simply the mean treatment effect of the group. This model also made promisingly accurate predictions in absolute terms: the correlation between empirical and predicted treatment response was 0.62 (95% CI 0.27, 0.95). Our results indicate that individuals' variation in response to anomia treatment are, at least somewhat, systematic and predictable, from the interaction between where and how much lesion damage they have suffered, and the time they devoted to the therapy.

Strategies for feature extraction from structural brain imaging in lesion-deficit modelling

High‐dimensional modelling of post‐stroke deficits from structural brain imaging is highly relevant to basic cognitive neuroscience and bears the potential to be translationally used to guide individual rehabilitation measures. One strategy to optimise model performance is well‐informed feature selection and representation. However, different feature representation strategies were so far used, and it is not known what strategy is best for modelling purposes. The present study compared the three common main strategies: voxel‐wise representation, lesion‐anatomical componential feature reduction and region‐wise atlas‐based feature representation. We used multivariate, machine‐learning‐based lesion‐deficit models to predict post‐stroke deficits based on structural lesion data. Support vector regression was tuned by nested cross‐validation techniques and tested on held‐out validation data to estimate model performance. While we consistently found the numerically best models for lower‐dimensional, featurised data and almost always for principal components extracted from lesion maps, our results indicate only minor, non‐significant differences between different feature representation styles. Hence, our findings demonstrate the general suitability of all three commonly applied feature representations in lesion‐deficit modelling. Likewise, model performance between qualitatively different popular brain atlases was not significantly different. Our findings also highlight potential minor benefits in individual fine‐tuning of feature representations and the challenge posed by the high, multifaceted complexity of lesion data, where lesion‐anatomical and functional criteria might suggest opposing solutions to feature reduction.