Interpreting and validating complexity and causality in lesion-symptom prognoses

Individualized lesion-symptom mapping using explainable artificial intelligence for the cognitive impact of white matter hyperintensities

Lesion-symptom mapping methods assess the relationship between lesions caused by cerebral small vessel disease and cognition, but current technology like support vector regression (SVR)) primarily provide group-level results. We propose a novel lesion-symptom mapping approach that can indicate how lesion patterns contribute to cognitive impairment on an individual level. A convolutional neural network (CNN) predicts cognitive scores and is combined with explainable artificial intelligence (XAI) to map the relation between cognition and vascular lesions. This method was evaluated primarily using real white matter hyperintensity maps of 821 memory clinic patients and simulated cognitive data, with weighted lesions and noise levels. Simulated data provided ground truth locations to assess predictive performance of the CNN and accuracy of strategic lesion identification by XAI, using an established lesion-symptom mapping method, SVR, and a simple fully connected neural network (FNN) as benchmarks. Real cognitive scores were used in a final proof-of-principle analysis. Predictive performance in simulation experiments was high for the CNN (R2 = 0.964), SVR (R2 = 0.875), and FNN (R2 = 0.863). CNN with XAI provided patient-specific attribution maps that highlighted the ground truth locations. All methods showed similar sensitivity to noise. Using real cognitive scores, SVR (R2 = 0.291) obtained a somewhat higher predictive performance than the CNN (R2 = 0.216), although both methods substantially exceeded the predictive performance of total WMH volume alone (R2 = 0.013). The FNN performed worse on real data (R2 = 0.020). To conclude, results show that CNNs combined with XAI can perform lesion-symptom mapping and generate individual attribution maps, which could be a valuable feature with further method development.

Symptomatology after damage to the angular gyrus through the lenses of modern lesion-symptom mapping

Brain-behavior relationships are complex. For instance, one might know a brain region's function(s) but still be unable to accurately predict deficit type or severity after damage to that region. Here, I discuss the case of damage to the angular gyrus (AG) that can cause left-right confusion, finger agnosia, attention deficit, and lexical agraphia, as well as impairment in sentence processing, episodic memory, number processing, and gesture imitation. Some of these symptoms are grouped under AG syndrome or Gerstmann's syndrome, though its exact underlying neuronal systems remain elusive. This review applies recent frameworks of brain-behavior modes and principles from modern lesion-symptom mapping to explain symptomatology after AG damage. It highlights four major issues for future studies: (1) functionally heterogeneous symptoms after AG damage need to be considered in terms of the degree of damage to (i) different subdivisions of the AG, (ii) different AG connectivity profiles that disconnect AG from distant regions, and (iii) lesion extent into neighboring regions damaged by the same infarct. (2) To explain why similar symptoms can also be observed after damage to other regions, AG damage needs to be studied in terms of the networks of regions that AG functions with, and other independent networks that might subsume the same functions. (3) To explain inter-patient variability on AG symptomatology, the degree of recovery-related brain reorganisation needs to account for time post-stroke, demographics, therapy input, and pre-stroke differences in functional anatomy. (4) A better integration of the results from lesion and functional neuroimaging investigations of AG function is required, with only the latter so far considering AG function in terms of a hub within the default mode network. Overall, this review discusses why it is so difficult to fully characterize the AG syndrome from lesion data, and how this might be addressed with modern lesion-symptom mapping.

Predicting recovery following stroke: Deep learning, multimodal data and feature selection using explainable AI

Machine learning offers great potential for automated prediction of post-stroke symptoms and their response to rehabilitation. Major challenges for this endeavour include the very high dimensionality of neuroimaging data, the relatively small size of the datasets available for learning and interpreting the predictive features, as well as, how to effectively combine neuroimaging and tabular data (e.g. demographic information and clinical characteristics). This paper evaluates several solutions based on two strategies. The first is to use 2D images that summarise MRI scans. The second is to select key features that improve classification accuracy. Additionally, we introduce the novel approach of training a convolutional neural network (CNN) on images that combine regions-of-interests (ROIs) extracted from MRIs, with symbolic representations of tabular data. We evaluate a series of CNN architectures (both 2D and a 3D) that are trained on different representations of MRI and tabular data, to predict whether a composite measure of post-stroke spoken picture description ability is in the aphasic or non-aphasic range. MRI and tabular data were acquired from 758 English speaking stroke survivors who participated in the PLORAS study. Each participant was assigned to one of five different groups that were matched for initial severity of symptoms, recovery time, left lesion size and the months or years post-stroke that spoken description scores were collected. Training and validation were carried out on the first four groups. The fifth (lock-box/test set) group was used to test how well model accuracy generalises to new (unseen) data. The classification accuracy for a baseline logistic regression was 0.678 based on lesion size alone, rising to 0.757 and 0.813 when initial symptom severity and recovery time were successively added. The highest classification accuracy (0.854), area under the curve (0.899) and F1 score (0.901) were observed when 8 regions of interest were extracted from each MRI scan and combined with lesion size, initial severity and recovery time in a 2D Residual Neural Network (ResNet). This was also the best model when data were limited to the 286 participants with moderate or severe initial aphasia (with area under curve = 0.865), a group that would be considered more difficult to classify. Our findings demonstrate how imaging and tabular data can be combined to achieve high post-stroke classification accuracy, even when the dataset is small in machine learning terms. We conclude by proposing how the current models could be improved to achieve even higher levels of accuracy using images from hospital scanners.

Stable multivariate lesion symptom mapping

Multivariate lesion-symptom mapping (MLSM) considers lesion information across the entire brain to predict impairments. The strength of this approach is also its weakness-considering many brain features together synergistically can uncover complex brain-behavior relationships but exposes a high-dimensional feature space that a model is expected to learn. Successfully distinguishing between features in this landscape can be difficult for models, particularly in the presence of irrelevant or redundant features. Here, we propose stable multivariate lesion-symptom mapping (sMLSM), which integrates the identification of reliable features with stability selection into conventional MLSM and describe our open-source MATLAB implementation. Usage is showcased with our publicly available dataset of chronic stroke survivors (N=167) and further validated in our independent public acute stroke dataset (N = 1106). We demonstrate that sMLSM eliminates inconsistent features highlighted by MLSM, reduces variation in feature weights, enables the model to learn more complex patterns of brain damage, and improves model accuracy for predicting aphasia severity in a way that tends to be robust regarding the choice of parameters for identifying reliable features. Critically, sMLSM more consistently outperforms predictions based on lesion size alone. This advantage is evident starting at modest sample sizes (N>75). Spatial distribution of feature importance is different in sMLSM, which highlights the features identified by univariate lesion symptom mapping while also implicating select regions emphasized by MLSM. Beyond improved prediction accuracy, sMLSM can offer deeper insight into reliable biomarkers of impairment, informing our understanding of neurobiology.

Consistent spatial lesion-symptom patterns: A comprehensive analysis using triangulation in lesion-symptom mapping in a cohort of stroke patients

INTRODUCTION

The relationship between brain lesions and stroke outcomes is crucial for advancing patient prognosis and developing effective therapies. Stroke is a leading cause of disability worldwide, and it is important to understand the neurological basis of its varied symptomatology. Lesion-symptom mapping (LSM) methods provide a means to identify brain areas that are strongly associated with specific symptoms. However, inner variations in LSM methods can yield different results. To address this, our study aimed to characterize the lesion-symptom mapping variability using three different LSM methods. Specifically, we sought to determine a lesion symptom core across LSM approaches enhancing the robustness of the analysis and removing potential spatial bias.

MATERIAL & METHODS

A cohort consisting of 35 patients with either right- or left-sided middle cerebral artery strokes were enrolled and evaluated using the NIHSS at 24 h post-stroke. Anatomical T1w MRI scans were also obtained 24 h post-stroke. Lesion masks were segmented manually and three distinctive LSM methods were implemented: ROI correlation-based, univariate, and multivariate approaches.

RESULTS

The results of the LSM analyses showed substantial spatial differences in the extension of each of the three lesion maps. However, upon overlaying all three lesion-symptom maps, a consistent lesion core emerged, corresponding to the territory associated with elevated NIHSS scores. This finding not only enhances the spatial accuracy of the lesion map but also underscores its clinical relevance.

CONCLUSION

This study underscores the significance of exploring complementary LSM approaches to investigate the association between brain lesions and stroke outcomes. By utilizing multiple methods, we can increase the robustness of our results, effectively addressing and neutralizing potential spatial bias introduced by each individual method. Such an approach holds promise for enhancing our understanding of stroke pathophysiology and optimizing patient care strategies.