Multi-Scale Deep Learning of Clinically Acquired Multi-Modal MRI Improves the Localization of Seizure Onset Zone in Children With Drug-Resistant Epilepsy.

The present study investigates the effectiveness of a deep learning neural network for non-invasively localizing the seizure onset zone (SOZ) using multi-modal MRI data that are clinically acquired from children with drug-resistant epilepsy. A cortical parcellation was applied to localize the SOZ in cortical nodes of the epileptogenic hemisphere. At each node, the laminar surface analysis was followed to sample 1) the relative intensity of gray matter and white matter in multi-modal MRI and 2) the neighboring white matter connectivity using diffusion tractography edge strengths. A cross-validation was employed to train and test all layers of a multi-scale residual neural network (msResNet) that can classify SOZ node in an end-to-end fashion. A prediction probability of a given node belonging to the SOZ class was proposed as a non-invasive MRI marker of seizure onset likelihood. In an independent validation cohort, the proposed MRI marker provided a very large effect size of Cohen's d = 1.21 between SOZ and non-SOZ, and classified SOZ with a balanced accuracy of 0.75 in lesional and 0.67 in non-lesional MRI groups. The subsequent multi-variate logistic regression found the incorporation of the proposed MRI marker into interictal intracranial EEG (iEEG) markers further improves the differentiation between the epileptogenic focus (defined as SOZ resected during surgery) and non-epileptogenic sites (i.e., non-SOZ sites preserved during surgery) up to 15 % in non-lesional MRI group, suggesting that the proposed MRI marker could improve the localization of epileptogenic foci for successful pediatric epilepsy surgery.

Developmental organization of neural dynamics supporting auditory perception.

PURPOSE: A prominent view of language acquisition involves learning to ignore irrelevant auditory signals through functional reorganization, enabling more efficient processing of relevant information. Yet, few studies have characterized the neural spatiotemporal dynamics supporting rapid detection and subsequent disregard of irrelevant auditory information, in the developing brain. To address this unknown, the present study modeled the developmental acquisition of cost-efficient neural dynamics for auditory processing, using intracranial electrocorticographic responses measured in individuals receiving standard-of-care treatment for drug-resistant, focal epilepsy. We also provided evidence demonstrating the maturation of an anterior-to-posterior functional division within the superior-temporal gyrus (STG), which is known to exist in the adult STG. METHODS: We studied 32 patients undergoing extraoperative electrocorticography (age range: eight months to 28 years) and analyzed 2,039 intracranial electrode sites outside the seizure onset zone, interictal spike-generating areas, and MRI lesions. Patients were given forward (normal) speech sounds, backward-played speech sounds, and signal-correlated noises during a task-free condition. We then quantified sound processing-related neural costs at given time windows using high-gamma amplitude at 70-110 Hz and animated the group-level high-gamma dynamics on a spatially normalized three-dimensional brain surface. Finally, we determined if age independently contributed to high-gamma dynamics across brain regions and time windows. RESULTS: Group-level analysis of noise-related neural costs in the STG revealed developmental enhancement of early high-gamma augmentation and diminution of delayed augmentation. Analysis of speech-related high-gamma activity demonstrated an anterior-to-posterior functional parcellation in the STG. The left anterior STG showed sustained augmentation throughout stimulus presentation, whereas the left posterior STG showed transient augmentation after stimulus onset. We found a double dissociation between the locations and developmental changes in speech sound-related high-gamma dynamics. Early left anterior STG high-gamma augmentation (i.e., within 200 ms post-stimulus onset) showed developmental enhancement, whereas delayed left posterior STG high-gamma augmentation declined with development. CONCLUSIONS: Our observations support the model that, with age, the human STG refines neural dynamics to rapidly detect and subsequently disregard uninformative acoustic noises. Our study also supports the notion that the anterior-to-posterior functional division within the left STG is gradually strengthened for efficient speech-sound perception after birth.

Diffusion tractography predicts propagated high-frequency activity during epileptic spasms.

OBJECTIVE: To determine the structural networks that constrain propagation of ictal oscillations during epileptic spasm events, and compare the observed propagation patterns across patients with successful or unsuccessful surgical outcomes. METHODS: Subdural electrode recordings of 18 young patients (age 1-11 years) were analyzed during epileptic spasm events to determine ictal networks and quantify the amplitude and onset time of ictal oscillations across the cortical surface. Corresponding structural networks were generated with diffusion magnetic resonance imaging (MRI) tractography by seeding the cortical region associated with the earliest average oscillation onset time, and white matter pathways connecting active electrode regions within the ictal network were isolated. Properties of this structural network were used to predict oscillation onset times and amplitudes, and this relationship was compared across patients who did and did not achieve seizure freedom following resective surgery. RESULTS: Onset propagation patterns were relatively consistent across each patient's spasm events. An electrode's average ictal oscillation onset latency was most significantly associated with the length of direct corticocortical tracts connecting to the area with the earliest average oscillation onset (p < .001, model R2  = .54). Moreover, patients demonstrating a faster propagation of ictal oscillation signals within the corticocortical network were more likely to have seizure recurrence following resective surgery (p = .039). In addition, ictal oscillation amplitude was associated with connecting tractography length and weighted fractional anisotropy (FA) measures along these pathways (p = .002/.030, model R2  = .31/.25). Characteristics of analogous corticothalamic pathways did not show significant associations with ictal oscillation onset latency or amplitude. SIGNIFICANCE: Spatiotemporal propagation patterns of high-frequency activity in epileptic spasms align with length and FA measures from onset-originating corticocortical pathways. Considering the data in this individualized framework may help inform surgical decision-making and expectations of surgical outcomes.

Deep Relational Reasoning for the Prediction of Language Impairment and Postoperative Seizure Outcome Using Preoperative DWI Connectome Data of Children With Focal Epilepsy.

Prolonged seizures in children with focal epilepsy (FE) may impair language functions and often reoccur after surgical intervention. This study is aimed at developing a novel deep relational reasoning network to investigate whether conventional diffusion-weighted imaging connectome analysis can be improved when predicting expressive and receptive scores of preoperative language impairments and classifying postoperative seizure outcomes (seizure freedom or recurrence) in individual FE children. To deeply reason the dependencies of axonal connections that are sparsely distributed in the whole brain, this study proposes the "dilated CNN + RN", a dilated convolutional neural network (CNN) combined with a relation network (RN). The performance of the dilated CNN + RN was evaluated using whole brain connectome data from 51 FE children. It was found that when compared with other state-of-the-art algorithms, the dilated CNN + RN led to an average improvement of 90.2% and 97.3% in predicting expressive and receptive language scores, and 2.2% and 4% improvement in classifying seizure freedom and seizure recurrence, respectively. These improvements were independent of the prefixed connectome densities. Also, the dilated CNN + RN could provide an explainable artificial intelligence (AI) model by computing gradient-based regression/classification activation maps. This mapping analysis revealed left superior-medial frontal cortex, bilateral hippocampi, and cerebellum as crucial hubs, facilitating important connections that were most predictive of language function and seizure refractoriness after surgery.

Novel Deep Learning Network Analysis of Electrical Stimulation Mapping-Driven Diffusion MRI Tractography to Improve Preoperative Evaluation of Pediatric Epilepsy.

OBJECTIVE: To investigate the clinical utility of deep convolutional neural network (DCNN) tract classification as a new imaging tool in the preoperative evaluation of children with focal epilepsy (FE). METHODS: A DCNN tract classification deeply learned spatial trajectories of DWI white matter pathways linking electrical stimulation mapping (ESM) findings from 89 children with FE, and then automatically identified white matter pathways associated with eloquent functions (i.e., primary motor, language, and vision). Clinical utility was examined by 1) measuring the nearest distance between DCNN-determined pathways and ESM, 2) evaluating the effectiveness of DCNN-determined pathways to optimize surgical margins via Kalman filter analysis, and 3) evaluating how accurately changes in DCNN-determined language pathway volume can predict changes in language ability via canonical correlation analysis. RESULTS: DCNN tract classification outperformed other existing methods, achieving an excellent accuracy of 98 % while non-invasively detecting eloquent areas within the spatial resolution of ESM (i.e., 1 cm). The Kalman filter analysis found that the preservation of brain areas within a surgical margin determined by DCNN tract classification predicted lack of postoperative deficit with a high accuracy of 92 %. Postoperative change of DCNN-determined language pathway volume showed a significant correlation with postoperative changes in language ability (R = 0.7, p   0.001). CONCLUSION: Our findings demonstrate that postoperative functional deficits substantially differ according to the extent of resected white matter, and that DCNN tract classification may offer key translational information by identifying these pathways in pediatric epilepsy surgery. SIGNIFICANCE: DCNN tract classification may be an effective tool to improve surgical outcome of children with FE.

Novel diffusion tractography methodology using Kalman filter prediction to improve preoperative benefit-risk analysis in pediatric epilepsy surgery.

OBJECTIVE: In this study the authors investigated the clinical reliability of diffusion weighted imaging maximum a posteriori probability (DWI-MAP) analysis with Kalman filter prediction in pediatric epilepsy surgery. This approach can yield a suggested resection margin as a dynamic variable based on preoperative DWI-MAP pathways. The authors sought to determine how well the suggested margin would have maximized occurrence of postoperative seizure freedom (benefit) and minimized occurrence of postoperative neurological deficits (risk). METHODS: The study included 77 pediatric patients with drug-resistant focal epilepsy (age 10.0 ± 4.9 years) who underwent resection of their presumed epileptogenic zone. In preoperative DWI tractography from the resected hemisphere, 9 axonal pathways, Ci=1-9, were identified using DWI-MAP as follows: C1-3 supporting face, hand, and leg motor areas; C4 connecting Broca's and Wernicke's areas; C5-8 connecting Broca's, Wernicke's, parietal, and premotor areas; and C9 connecting the occipital lobe and lateral geniculate nucleus. For each Ci, the resection margin, di, was measured by the minimal Euclidean distance between the voxels of Ci and the resection boundary determined by spatially coregistered postoperative MRI. If Ci was resected, di was assumed to be negative (calculated as -1 × average Euclidean distance between every voxel inside the resected Ci volume, ri). Kalman filter prediction was then used to estimate an optimal resection margin, d*i, to balance benefit and risk by approximating the relationship between di and ri. Finally, the authors defined the preservation zone of Ci that can balance the probability of benefit and risk by expanding the cortical area of Ci up to d*i on the 3D cortical surface. RESULTS: In the whole group (n = 77), nonresection of the preoperative preservation zone (i.e., actual resection margin d*i greater than the Kalman filter-defined d*i) accurately predicted the absence of postoperative motor (d*1-3: 0.93 at seizure-free probability of 0.80), language (d*4-8: 0.91 at seizure-free probability of 0.81), and visual deficits (d*9: 0.90 at seizure-free probability of 0.75), suggesting that the preservation of preoperative Ci within d*i supports a balance between postoperative functional deficit and seizure freedom. The subsequent subgroup analyses found that preservation of preoperative Ci =1-4,9 within d*i =1-4,9 may provide accurate deficit predictions independent of age and seizure frequency, suggesting that the DWI-based surgical margin can be effective for surgical planning even in young children and across a range of epilepsy severity. CONCLUSIONS: Integrating DWI-MAP analysis with Kalman filter prediction may help guide epilepsy surgery by visualizing the margins of the eloquent white matter pathways to be preserved.

Prediction of postoperative deficits using an improved diffusion-weighted imaging maximum a posteriori probability analysis in pediatric epilepsy surgery.

OBJECTIVEThis study is aimed at improving the clinical utility of diffusion-weighted imaging maximum a posteriori probability (DWI-MAP) analysis, which has been reported to be useful for predicting postoperative motor, language, and visual field deficits in pediatric epilepsy surgery. The authors determined the additive value of a new clustering mapping method in which average direct-flip distance (ADFD) reclassifies the outliers of original DWI-MAP streamlines by referring to their minimum distances to the exemplar streamlines (i.e., medoids).METHODSThe authors studied 40 children with drug-resistant focal epilepsy (mean age 8.7 ± 4.8 years) who had undergone resection of the presumed epileptogenic zone and had five categories of postoperative deficits (i.e., hemiparesis involving the face, hand, and/or leg; dysphasia requiring speech therapy; and/or visual field cut). In pre- and postoperative images of the resected hemisphere, DWI-MAP identified a total of nine streamline pathways: C1 = face motor area, C2 = hand motor area, C3 = leg motor area, C4 = Broca's area-Wernicke's area, C5 = premotor area-Broca's area, C6 = premotor area-Wernicke's area, C7 = parietal area-Wernicke's area, C8 = premotor area-parietal area, and C9 = occipital lobe-lateral geniculate nucleus. For each streamline of the identified pathway, the minimal ADFD to the nine exemplars corrected the pathway membership. Binary logistic regression analysis was employed to determine how accurately two fractional predictors, Δ1-9 (postoperative volume change of C1-9) and γ1-9 (preoperatively planned volume of C1-9 resected), predicted postoperative motor, language, and visual deficits.RESULTSThe addition of ADFD to DWI-MAP analysis improved the sensitivity and specificity of regression models for predicting postoperative motor, language, and visual deficits by 28% for Δ1-3 (from 0.62 to 0.79), 13% for Δ4-8 (from 0.69 to 0.78), 13% for Δ9 (from 0.77 to 0.87), 7% for γ1-3 (from 0.81 to 0.87), 1% for γ4-8 (from 0.86 to 0.87), and 24% for γ9 (from 0.75 to 0.93). Preservation of the eloquent pathways defined by preoperative DWI-MAP analysis with ADFD (up to 97% of C1-4,9) prevented postoperative motor, language, and visual deficits with sensitivity and specificity ranging from 88% to 100%.CONCLUSIONSThe present study suggests that postoperative functional outcome substantially differs according to the extent of resected white matter encompassing eloquent cortex as determined by preoperative DWI-MAP analysis. The preservation of preoperative DWI-MAP-defined pathways may be crucial to prevent postoperative deficits. The improved DWI-MAP analysis may provide a complementary noninvasive tool capable of guiding the surgical margin to minimize the risk of postoperative deficits for children.