Poster D19, Wednesday, August 21, 2019, 5:15 – 7:00 pm, Restaurant Hall
Establishing brain-to-behaviour prediction models of post-stroke aphasia: A systematic investigation of brain parcellations, multimodal imaging, and machine learning algorithms.
Ajay Halai1, Anna Woollams2, Matthew Lambon Ralph1;1MRC Cognition and Brain Sciences Unit, University of Cambridge, 2Neuroscience and Aphasia Research Unit, Faculty of Biology, Medicine and Health, University of Manchester
In recent decades, structural and functional neuroimaging have radically improved our understanding of how speech and language abilities map to the brain in normal and impaired participants, including the diverse, graded variations observed in post-stroke aphasia. Despite the potential for a paradigm shift in neuroscience theory and clinical practice, only recently have a handful of studies begun to explore the reverse inference: creating brain-to-behaviour prediction models. In order to establish some key foundations for successful prediction models, in this large-scale study we systematically investigated four critical issues to determine the optimal: (1) behavioural measures to use as targets; (2) partitioning of the brain space for use as predictive features; (3) combination of structural and connectivity measures from multimodal neuroimaging; and (4) type of machine learning algorithms to generate predictions. There is increasing agreement that binary aphasia classifications are limited; furthermore, while it is possible to predict performance on individual neuropsychological tests, any assessment taps multiple underlying component abilities and hence performance across tests is often intercorrelated over patients. An alternative approach places patients as points in a continuous multidimensional space, where the axes represent primary neuro-computational processes. Therefore, we explored the influence of the core model factors while predicting four principal dimensions of language and cognition variation in post-stroke aphasia. The results showed that across all four behavioural dimensions, we consistently found that the best prediction models were derived from structural measures extracted from T1 scans, parcellated using a larger number of ROIs and submitted to a multi-kernel learning algorithm. Adding information on white matter connectivity (in vivo patient-specific diffusion weighted data) did not improve the models. Our results provide a set of principles to guide future work aiming to predict outcomes in language disorders from brain imaging data. From a clinical implementation perspective, it suggests that the majority of the information needed for effective outcome prediction in stroke aphasia can be obtained using standard clinical MR scanning protocols, obviating the need for acquisition and processing of diffusion weighted data.
Themes: Disorders: Acquired, Methods
Method: Other