You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Copy file name to clipboardExpand all lines: src/Components/Main.jsx
+8-9Lines changed: 8 additions & 9 deletions
Original file line number
Diff line number
Diff line change
@@ -48,7 +48,7 @@ function Main() {
48
48
Cortical sites critical to language function act as connectors between language subnetworks
49
49
</h2>
50
50
<p>
51
-
Historically, eloquent functions have been viewed as localized to focal areas of human cerebral cortex, while more recent studies suggest they are encoded by distributed networks. We examined the network properties of cortical sites defined by stimulation to be critical for speech and language, using electrocorticography from sixteen participants during word-reading. We discovered distinct network signatures for sites where stimulation caused speech arrest and language errors. Both demonstrated lower local and global connectivity, whereas sites causing language errors exhibited higher inter-community connectivity, identifying them as connectors between modules in the language network. We used machine learning to classify these site types with reasonably high accuracy, even across participants, suggesting that a site’s pattern of connections within the task-activated language network helps determine its importance to function. These findings help to bridge the gap in our understanding of how focal cortical stimulation interacts with complex brain networks to elicit language deficits.
51
+
Historically, eloquent functions have been viewed as localized to focal areas of human cerebral cortex, while more recent studies suggest they are encoded by distributed networks. We examined the network properties of cortical sites defined by stimulation to be critical for speech and language, using electrocorticography from sixteen participants during word-reading. We discovered distinct network signatures for sites where stimulation caused speech arrest and language errors. Both demonstrated lower local and global connectivity, whereas sites causing language errors exhibited higher inter-community connectivity, identifying them as connectors between modules in the language network. We used machine learning to classify these site types with reasonably high accuracy, even across participants, suggesting that a site’s pattern of connections within the task-activated language network helps determine its importance to function. These findings help to bridge the gap in our understanding of how focal cortical stimulation interacts with complex brain networks to elicit language deficits.
"A DES was used either intraoperatively (depicted) or in the epilepsy monitoring unit to identify sites critical to language and speech. These were subdivided into cortical regions causing language errors (LE) or speech arrest (SA). B We recorded continuous ECoG while participants engaged in a word-reading task. C We generated one static network for each participant using pairwise high-gamma correlations. Color-coded adjacency matrix shown; the color in position (m,n) reflects to the high-gamma correlation between electrode m and n. r is the Fisher-transformed Pearson correlation. Community partitions were discovered using modularity maximization. Electrodes have been re-ordered so those belonging to the same community are adjacent (boundaries shown in black lines). D Spring-loaded network plot; nodes (circles) that are more strongly connected are drawn more closely together. The size of each node is proportional to its strength. Community membership is indicated by the fill color of each node. The nodes outlined in blue are LE nodes. E Network metrics were calculated—PC (participation coefficient), strength, CC (clustering coefficient), LE (local efficiency), and EC (eigenvector centrality). Metric values for every node are plotted; large colored points represent critical nodes and small gray points are all other nodes. Boxes demonstrate the median and interquartile range. We used these metrics to train machine learning classifiers to predict which nodes would be critical to language and speech. Example data (C–E) are provided from a single participant (n = 1) for each visualization. Source data are provided as a Source Data file."
81
-
</p>
82
-
)}
79
+
<p>
80
+
"A DES was used either intraoperatively (depicted) or in the epilepsy monitoring unit to identify sites critical to language and speech. These were subdivided into cortical regions causing language errors (LE) or speech arrest (SA). B We recorded continuous ECoG while participants engaged in a word-reading task. C We generated one static network for each participant using pairwise high-gamma correlations. Color-coded adjacency matrix shown; the color in position (m,n) reflects to the high-gamma correlation between electrode m and n. r is the Fisher-transformed Pearson correlation. Community partitions were discovered using modularity maximization. Electrodes have been re-ordered so those belonging to the same community are adjacent (boundaries shown in black lines). D Spring-loaded network plot; nodes (circles) that are more strongly connected are drawn more closely together. The size of each node is proportional to its strength. Community membership is indicated by the fill color of each node. The nodes outlined in blue are LE nodes. E Network metrics were calculated—PC (participation coefficient), strength, CC (clustering coefficient), LE (local efficiency), and EC (eigenvector centrality). Metric values for every node are plotted; large colored points represent critical nodes and small gray points are all other nodes. Boxes demonstrate the median and interquartile range. We used these metrics to train machine learning classifiers to predict which nodes would be critical to language and speech. Example data (C–E) are provided from a single participant (n = 1) for each visualization. Source data are provided as a Source Data file."</p>
81
+
)}
83
82
{index1===1&&(
84
83
<p>
85
-
"PC participation coefficient, S strength, CC clustering coefficient, LEff local efficiency, EC eigenvector centrality. A Diagram illustrating coassignment. Two yellow-outlined coassigned nodes are found within the same community (dark blue fill); two blue-outlined nodes are found in two different communities (magenta and orange fill)—i.e., not coassigned. B Diagram demonstrating graph metrics. The large magenta node in the top panel has a high PC—it connects across all communities in this network. The same node has a low clustering coefficient (its neighbors are not themselves connected, denoted by dashed arrows) and low local efficiency (long path lengths between its neighbors). In the bottom panel, the large dark blue node has high strength, i.e., a high sum of connection weights. The large orange node has higher eigenvector centrality than the smaller orange node; both have the same number of connections, but the larger node’s connections themselves have more connections. C Intuition for three node types. Connector nodes connect across communities (high PC), while their neighbors do not connect as closely to each other (low CC, LEff). Global hubs connect to many nodes across the network (high PC, high S, likely high EC). Local hubs connect densely in their neighborhood (low PC, high CC/LEff)." </p>
84
+
"PC participation coefficient, S strength, CC clustering coefficient, LEff local efficiency, EC eigenvector centrality. A Diagram illustrating coassignment. Two yellow-outlined coassigned nodes are found within the same community (dark blue fill); two blue-outlined nodes are found in two different communities (magenta and orange fill)—i.e., not coassigned. B Diagram demonstrating graph metrics. The large magenta node in the top panel has a high PC—it connects across all communities in this network. The same node has a low clustering coefficient (its neighbors are not themselves connected, denoted by dashed arrows) and low local efficiency (long path lengths between its neighbors). In the bottom panel, the large dark blue node has high strength, i.e., a high sum of connection weights. The large orange node has higher eigenvector centrality than the smaller orange node; both have the same number of connections, but the larger node’s connections themselves have more connections. C Intuition for three node types. Connector nodes connect across communities (high PC), while their neighbors do not connect as closely to each other (low CC, LEff). Global hubs connect to many nodes across the network (high PC, high S, likely high EC). Local hubs connect densely in their neighborhood (low PC, high CC/LEff)."</p>
86
85
)}
87
86
{index1===2&&(
88
87
<p>
89
-
"PC participation coefficient, S strength, CC clustering coefficient, LEff local efficiency, EC eigenvector centrality. A Composite of all participants’ electrodes colocalized on a single template brain. Speech arrest nodes (yellow fill) were primarily located in ventral premotor regions, but also in ventrolateral prefrontal and ventral temporal regions. Language error nodes (blue fill) were widely distributed in perisylvian regions. B Three example participant brain reconstructions. Node color (filled) represents community assignment, and node size is proportional to its participation coefficient. The outline color indicates critical nodes (blue—LE node, yellow—SA node). C Corresponding three network diagrams. The electrode position is spring-weighted (stronger connections draw electrodes closer together). Fill color indicates community, and if present, outline color indicates critical node type (LE vs. SA) D Corresponding network metrics for the three example patients. Metrics for all nodes (electrodes) for each of the three participants (n = 1 per graph) are plotted. Here, colored circles represent critical nodes; gray circles represent other nodes. Boxes demonstrate median and interquartile range, and whiskers demonstrate non-outlier maxima/minima. Source data are provided as a Source Data file." </p>
88
+
"PC participation coefficient, S strength, CC clustering coefficient, LEff local efficiency, EC eigenvector centrality. A Composite of all participants’ electrodes colocalized on a single template brain. Speech arrest nodes (yellow fill) were primarily located in ventral premotor regions, but also in ventrolateral prefrontal and ventral temporal regions. Language error nodes (blue fill) were widely distributed in perisylvian regions. B Three example participant brain reconstructions. Node color (filled) represents community assignment, and node size is proportional to its participation coefficient. The outline color indicates critical nodes (blue—LE node, yellow—SA node). C Corresponding three network diagrams. The electrode position is spring-weighted (stronger connections draw electrodes closer together). Fill color indicates community, and if present, outline color indicates critical node type (LE vs. SA) D Corresponding network metrics for the three example patients. Metrics for all nodes (electrodes) for each of the three participants (n = 1 per graph) are plotted. Here, colored circles represent critical nodes; gray circles represent other nodes. Boxes demonstrate median and interquartile range, and whiskers demonstrate non-outlier maxima/minima. Source data are provided as a Source Data file." </p>
90
89
)}
91
90
{index1===3&&(
92
91
<p>
93
-
"PC participation coefficient, S strength, CC clustering coefficient, LEff local efficiency, EC eigenvector centrality. *p < 0.05. **p < 0.01. ***p < 0.001 (FDR-corrected). A Histogram of the number of communities per participant (n = 16). B Coassignment percentages vs. chance. Coassignment is calculated as the mean % of critical, LE, or SA node pairs per participant sharing a community. Empiric chance was calculated based on 1000 random shuffles of community assignment per participant, presented as mean coassignment% per participant with bars indicating standard error of mean (n = 16 for Critical, n = 15 for LE and SA). Critical nodes, language error nodes, and speech arrest nodes were significantly more likely to coassign in the same communities than chance (p < 0.001 for all, one-tailed estimate against empiric chance). Language error and speech arrest nodes were not more likely to be found in the same community as each other compared to chance (35.2 vs. 30.4%, p = 0.112, one-tailed estimate against empiric chance). C Network metrics for critical vs. all other nodes (150 critical nodes, 1084 non-critical nodes). Critical nodes have higher PC and lower CC, LEff, and EC than other nodes. D Network metrics for LE, SA, and other nodes (92 language error nodes, 52 speech arrest nodes, 1084 non-critical nodes). LE nodes have markedly higher PC than SA and other nodes. C, D Metrics were z-scored for each subject prior to pooling all nodes together. All nodes are plotted in light gray; mean values per participant in larger, bolder colors. Boxes indicate the median and IQR, and notch indicates the standard error of the median. Statistical testing is based on a two-sided two-sample t-test on z-scored metrics across all pooled nodes with FDR correction. For additional details, refer to Table 1. Source data are provided as a Source Data file." </p>
92
+
"PC participation coefficient, S strength, CC clustering coefficient, LEff local efficiency, EC eigenvector centrality. *p < 0.05. **p < 0.01. ***p < 0.001 (FDR-corrected). A Histogram of the number of communities per participant (n = 16). B Coassignment percentages vs. chance. Coassignment is calculated as the mean % of critical, LE, or SA node pairs per participant sharing a community. Empiric chance was calculated based on 1000 random shuffles of community assignment per participant, presented as mean coassignment% per participant with bars indicating standard error of mean (n = 16 for Critical, n = 15 for LE and SA). Critical nodes, language error nodes, and speech arrest nodes were significantly more likely to coassign in the same communities than chance (p < 0.001 for all, one-tailed estimate against empiric chance). Language error and speech arrest nodes were not more likely to be found in the same community as each other compared to chance (35.2 vs. 30.4%, p = 0.112, one-tailed estimate against empiric chance). C Network metrics for critical vs. all other nodes (150 critical nodes, 1084 non-critical nodes). Critical nodes have higher PC and lower CC, LEff, and EC than other nodes. D Network metrics for LE, SA, and other nodes (92 language error nodes, 52 speech arrest nodes, 1084 non-critical nodes). LE nodes have markedly higher PC than SA and other nodes. C, D Metrics were z-scored for each subject prior to pooling all nodes together. All nodes are plotted in light gray; mean values per participant in larger, bolder colors. Boxes indicate the median and IQR, and notch indicates the standard error of the median. Statistical testing is based on a two-sided two-sample t-test on z-scored metrics across all pooled nodes with FDR correction. For additional details, refer to Table 1. Source data are provided as a Source Data file."</p>
94
93
)}
95
94
{index1===4&&(
96
95
<p>
97
-
"For within-participant classification, participants with at least four nodes of the relevant class were included; for critical nodes, LE nodes, and SA nodes, n = 15, 10, and 8, respectively. For across-participant classification, participants with at least one node of the relevant class were included—for critical nodes, LE nodes, and SA nodes, n = 16, 13, and 13, respectively. A–D Each dot represents average classification balanced accuracy or sensitivity for a single participant. Box plots show median and IQR across participants and are derived from a single value per participant. Whiskers indicate a non-outlier maximum range. True balanced accuracy and sensitivity were compared against empirical chance calculated by label-shuffling. The average chance classification accuracy per participant is represented by the chance box plots for SVN and KNN (one value per participant). Data for SVM, KNN, and chance for SVM and KNN are presented in different colors as indicated by the legend. E, F ROC curves presented for SVM (solid lines) and KNN (dashed lines) classifiers, when classifying SA (orange), LE (magenta), and critical (dark blue) nodes separately, as indicated by the legend. For further details, refer to Tables 2, 3. Source data are provided as a Source Data file." </p>
96
+
"For within-participant classification, participants with at least four nodes of the relevant class were included; for critical nodes, LE nodes, and SA nodes, n = 15, 10, and 8, respectively. For across-participant classification, participants with at least one node of the relevant class were included—for critical nodes, LE nodes, and SA nodes, n = 16, 13, and 13, respectively. A–D Each dot represents average classification balanced accuracy or sensitivity for a single participant. Box plots show median and IQR across participants and are derived from a single value per participant. Whiskers indicate a non-outlier maximum range. True balanced accuracy and sensitivity were compared against empirical chance calculated by label-shuffling. The average chance classification accuracy per participant is represented by the chance box plots for SVN and KNN (one value per participant). Data for SVM, KNN, and chance for SVM and KNN are presented in different colors as indicated by the legend. E, F ROC curves presented for SVM (solid lines) and KNN (dashed lines) classifiers, when classifying SA (orange), LE (magenta), and critical (dark blue) nodes separately, as indicated by the legend. For further details, refer to Tables 2, 3. Source data are provided as a Source Data file."</p>
0 commit comments