Search

Reorganization of the language network from typically left-lateralized frontotemporal regions to bilaterally distributed or right-lateralized networks occurs in anywhere from 25%-30% of patients with focal epilepsy. In patients who have been recently diagnosed with epilepsy, an important question remains as to whether it is the presence of seizures or the underlying epilepsy etiology that leads to atypical language representations. This question becomes even more interesting in pediatric samples, where the typical developmental processes of the language network may confer more variability and plasticity in the language network. We assessed a carefully selected cohort of children with recent-onset epilepsy to examine whether it is the effects of seizures or their underlying cause that leads to atypical language lateralization.We used functional magnetic resonance imaging (fMRI) to compare language laterality in children with recently diagnosed focal unaware epilepsy and age-matched controls. Age at epilepsy onset (age 4 to 6 years vs age 7 to 12 years) was also examined to determine if age at onset influenced laterality.The majority of recent-onset patients and controls exhibited left-lateralized language. There was a significant interaction such that the relationship between epilepsy duration and laterality differed by age at onset. In children with onset after age 6, a longer duration of epilepsy was associated with less left-lateralized language dominance. In contrast, in children with onset between 4 and 6 years of age, a longer duration of epilepsy was not associated with less left language dominance.Our results demonstrate that although language remained largely left-lateralized in children recently diagnosed with epilepsy, the impact of seizure duration depended on age at onset, indicating that the timing of developmental and disease factors are important in determining language dominance.

One outstanding challenge for machine learning in diagnostic biomedical imaging is algorithm interpretability. A key application is the identification of subtle epileptogenic focal cortical dysplasias (FCDs) from structural MRI. FCDs are difficult to visualize on structural MRI but are often amenable to surgical resection. We aimed to develop an open-source, interpretable, surface-based machine-learning algorithm to automatically identify FCDs on heterogeneous structural MRI data from epilepsy surgery centres worldwide. The Multi-centre Epilepsy Lesion Detection (MELD) Project collated and harmonized a retrospective MRI cohort of 1015 participants, 618 patients with focal FCD-related epilepsy and 397 controls, from 22 epilepsy centres worldwide. We created a neural network for FCD detection based on 33 surface-based features. The network was trained and cross-validated on 50% of the total cohort and tested on the remaining 50% as well as on 2 independent test sites. Multidimensional feature analysis and integrated gradient saliencies were used to interrogate network performance. Our pipeline outputs individual patient reports, which identify the location of predicted lesions, alongside their imaging features and relative saliency to the classifier. On a restricted ‘gold-standard’ subcohort of seizure-free patients with FCD type IIB who had T1 and fluid-attenuated inversion recovery MRI data, the MELD FCD surface-based algorithm had a sensitivity of 85%. Across the entire withheld test cohort the sensitivity was 59% and specificity was 54%. After including a border zone around lesions, to account for uncertainty around the borders of manually delineated lesion masks, the sensitivity was 67%. This multicentre, multinational study with open access protocols and code has developed a robust and interpretable machine-learning algorithm for automated detection of focal cortical dysplasias, giving physicians greater confidence in the identification of subtle MRI lesions in individuals with epilepsy.

IntroductionOne outstanding challenge for machine learning in diagnostic biomedical imaging is algorithm interpretability. A key application is the identification of subtle epileptogenic focal cortical dysplasias (FCDs) from structural MRI. FCDs are difficult to visualise on structural MRI but are often amenable to surgical resection. We aimed to develop an open-source, interpretable, surface-based machine-learning algorithm to automatically identify FCDs on heterogeneous structural MRI data from epilepsy surgery centres worldwide.MethodsThe Multi-centre Epilepsy Lesion Detection (MELD) Project collated and harmonised a retrospective MRI cohort of 1015 participants, 618 patients with focal FCD-related epilepsy and 397 controls, from 22 epilepsy centres worldwide. We created a neural network for FCD detection based on 33 surface-based features. The network was trained and cross-validated on 50% of the total cohort and tested on the remaining 50% as well as on 2 independent test sites. Multidimensional feature analysis and integrated gradient saliencies were used to interrogate network performance.ResultsOur pipeline outputs individual patient reports, which identify the location of predicted lesions, alongside their imaging features and relative saliency to the classifier. Overall, after including a border-zone around lesions, the developed MELD FCD surface-based algorithm had a sensitivity of 67% and a specificity of 54% on the withheld test cohort, and a sensitivity of 85% on a restricted subcohort of seizure free patients with FCD type IIB who had T1 and FLAIR MRI data.ConclusionsThis multicentre, multinational study with open access protocols and code has developed a robust and interpretable machine-learning algorithm for automated detection of focal cortical dysplasias, giving physicians greater confidence in the identification of subtle MRI lesions.HighlightsThis large, multi-centre, multi-scanner neuroimaging cohort captures the heterogeneity of histopathological subtypes and imaging features of patients with FCD.We developed a robust and interpretable surface-based algorithm which detects FCDs with a sensitivity of 67% and a specificity of 54%.The algorithm generates individual patient reports that “open the AI black-box” highlighting predicted lesion locations, alongside the imaging features and their relative saliency to the classifier.

The Common Data Element (CDE) Project was initiated in 2006 by the National Institute of Neurological Disorders and Stroke (NINDS) to develop standards for performing funded neuroscience-related clinical research. CDEs are intended to standardize aspects of data collection, decrease study start-up time, and provide more complete, comprehensive, and equivalent data across studies within a particular disease area. Therefore, CDEs will simplify data sharing and data aggregation across NINDS-funded clinical research, and where appropriate, facilitate the development of evidenced-based guidelines and recommendations. Epilepsy-specific CDEs were established in nine content areas: (1) Antiepileptic Drugs (AEDs) and Other Antepileptic Therapies (AETs), )2) Comorbidities, (3) Electrophysiology, (4) Imaging,(5) Neurological Exam, (6) Neuropsychology,(7) Quality of Life, (8) Seizures and Syndromes, and (9) Surgery and Pathology. CDEs were developed as a dynamic resource that will accommodate recommendations based on investigator use, new technologies, and research findings documenting emerging critical disease characteristics. The epilepsy-specific CDE initiative can be viewed as part of the larger international movement toward “harmonization” of clinical disease characterization and outcome assessment designed to promote communication and research efforts in epilepsy. It will also provide valuable guidance for CDE improvement during their further development, refinement, and implementation. This article describes the NINDS CDE Initiative, the process used in developing Epilepsy CDEs, and the benefits of CDEs for the clinical investigator and NINDS.

The extent to which visual word perception engages speech codes (i.e., phonological recoding) remains a crucial question in understanding mechanisms of reading. In this study, we used functional magnetic resonance imaging techniques combined with behavioral response measures to examine neural responses to focused versus incidental phonological and semantic processing of written words. Three groups of subjects made simple button-pressing responses in either phonologically (rhyming-judgment) or semantically (category-judgment) focused tasks or both tasks with identical sets of visual stimuli. In the phonological tasks, subjects were given both words and pseudowords separated in different scan runs. The baseline task required feature search of scrambled letter strings created from the stimuli for the experimental conditions. The results showed that cortical regions associated with both semantic and phonological processes were strongly activated when the task required active processing of word meaning. However, when subjects were actively processing the speech sounds of the same set of written words, brain areas typically engaged in semantic processing became silent. In addition, subjects who performed both the rhyming and the semantic tasks showed diverse and significant bilateral activation in the prefrontal, temporal, and other brain regions. Taken together, the pattern of brain activity provides evidence of a neural basis supporting the theory that in normal word reading, phonological recoding is automatic and facilitates semantic processing of written words, while rapid comprehension of word meaning requires devoted attention. These results also raise questions about including multiple cognitive tasks in the same neuroimaging sessions.