Potent DWI marker to predict the improvement of core language after pediatric epilepsy surgery
Poster No:
363
Submission Type:
Abstract Submission
Authors:
Jeong-Won Jeong1, Min-Hee Lee1, Hiroshi Uda1, Aimee Luat1, Csaba Juhasz1, Eishi Asano1
Institutions:
1Wayne State University, Detroit, MI
First Author:
Co-Author(s):
Introduction:
Subsets of children with drug-resistant focal epilepsy show improvement in language skills postoperatively, which may enhance the overall quality of life for these individuals1,2. This study seeks to explore a preoperative imaging marker specific to children with left-hemispheric seizure focus who had a short-term postoperative language enhancement. We hypothesized that the right hemisphere of such patients would have an increase in "local efficiency of axonal connectivity" that facilitates information transfer between brain regions, especially associated with core language function.
Methods:
19 patients with drug-resistant epilepsy associated with left-hemispheric seizure focus and language dominance (11.9±4.3 years old) underwent both preoperative 3T DWI tractography scan using 55 encoding directions at b=1000 s/mm2 and pre-/postoperative neuropsychological language tests (average interval: 2.5 months, 6 patients improved core language scores after surgery). 3T DWI tractography data of 28 age-matched healthy controls were also obtained at the same scanner. Whole-brain tractography was sorted to construct two whole-brain backbone DWI connectomes (DWIC), whose elements were total tract counts of pair-wise connections weighted by average fractional anisotropy (FA) values to account for the axonal integrity inferring alterations in the axonal diameter, fiber density or myelin structure, 1) raw DWIC with a total of 1477 connections that may have false-positive tracts3 in each connection (i.e., wiggly tract and broken tract, etc.) and 2) clean DWIC with the 1477 connections, where deep convolutional neural network (DCNN) tract classification3 removes all potential false-positive tracts in each connection. From each DWIC of two DWIC data, left and right intra-hemispheric networks were extracted. In each network, local efficiency analogue (LEA)4 related to the average resistance distance between a given node and the remaining nodes was calculated as a metric quantifying the efficiency of local information flow from the given node. Lateralization index (i.e., LI = [right LEA–left LEA]/[right LEA+left LEA]) was extracted from individual nodes. The Kruskal-Wallis test was then used to identify key nodes showing significant difference of LI value between two groups: improvement vs. no improvement. The LI values of key nodes were fused using ranked supervised multivariate canonical correlation (SMVCCA)5 and evaluated using a multi-layer perceptron (MLP) to predict patients with postsurgical language improvement via 3-fold cross validation.
Results:
In contrast to the no improvement group, the improvement group exhibited higher LI values across multiple nodes of the intra-hemispheric network with significant group differences at p ≤ 0.05 (Fig. 1). Notably, the LI values in the improvement group surpassed even those of healthy controls. Clean DWIC demonstrated superior accuracy in predicting three classes, with a BA of 84±9% from MLP and BA of 93±3% from MLP with ranked sMVCCA, compared to raw DWIC (68±9% and 85±2%, left plot of Fig. 2). The LI values at multiple nodes, including thalamus (THA), superior temporal gyrus (STG), precentral gyrus (preCG), superior parietal gyrus (SPG), and superior occipital gyrus (SOC), were identified as crucial markers that exhibited atypical increases of LEA values (and FA values) in the right hemisphere, contributing significantly to the prediction of short-term language improvement (right plot of Fig. 2).
Conclusions:
This study provided initial evidence in left-hemispheric epilepsy indicating an increase in local efficiency and axonal integrity within the contralateral motor-language-visual network that could be a specific pattern of neural plasticity influencing the favorable likelihood of postoperative language improvement. More investigation with a larger study cohort is essential to identify who is likely to benefit from early surgery.
Disorders of the Nervous System:
Neurodevelopmental/ Early Life (eg. ADHD, autism) 1
Language:
Language Other
Modeling and Analysis Methods:
Connectivity (eg. functional, effective, structural) 2
Novel Imaging Acquisition Methods:
Diffusion MRI
Keywords:
Epilepsy
Language
Pediatric Disorders
Plasticity
Tractography
White Matter
1|2Indicates the priority used for review
Provide references using author date format
[2] Edwards, J. C., (2018), “Marginal decision-making in the treatment of refractory epilepsy”, J. Med. Econ., vol. 21, no. 5, pp. 438–442.
[3] Lee, M.-H., (2019), “Improving reproducibility of diffusion connectome analysis using deep convolutional neural network model”, Proc. Int. Soc. Magn. Reson. Med., p. 3578.
[4] Klein, D. J. (1993), “Resistance distance”, Journal of Mathematical Chemistry, vol. 12, pp. 81–95, 1993.
[5] Lee, G., (2015), “Supervised multi-view canonical correlation analysis (smvcca): Integrating histologic and proteomic features for predicting recurrent prostate cancer”, IEEE Trans. Med. Imaging, vol. 34, no. 1, pp. 284–297.