Lesion Normalization and Supervised Learning in Post-traumatic Seizure Classification with Diffusion MRI

Akbar, Md Navid; Ruf, Sebastian F.; Rocca, Marianna La; Garner, Rachael; Barisano, Giuseppe; Cua, Ruskin; Vespa, Paul; Erdoğmuş, Deni̇z; Duncan, Dominique

doi:10.1007/978-3-030-87615-9_12

Cited by 1 publication

(8 citation statements)

References 24 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Aligned with our finding in Table IV, inferior temporal gyrus has been previously reported as a brain biomarker for early PTS (EPTS) and LPTS, following shape analysis [44]. It is also worth noting that alterations in the other regions, which we found to be statistically significant before the Bonferroni correction, have also been shown to be associated with PTSrelated abnormalities: fornix/stria terminalis (FX/ST) [16], external capsule (EC) [45], and lingual gyri (regions-47,48) [46]. There is also the possibility that other potential biomarkers did not demonstrate a significant group level difference, owing to their variations within the individual level [47].…”

Section: Discussionsupporting

confidence: 85%

“…8). Abnormalities of dMRI in ALIC-R (ALIC right) have been documented earlier in subjects suffering from LPTS (Akbar et al, 2021), and that in ALIC (both left and right) in temporal lobe epilepsy (TLE) (Meng et al, 2010). ALIC, or more precisely ALIC left, is thus our strongest candidate for a potential biomarker.…”

Section: Discussionmentioning

confidence: 75%

“…In all classification tasks, whether as an interim individual learner or as the final estimator, both SVM and AdaBoost classifiers were investigated. The choice of these two classifiers was aligned with previous studies, which have found both SVM [34], [35] and AdaBoost [16] to be suitable for classification tasks involving neuroimaging features. For SVM, both linear and radial basis function (RBF) kernel variants were explored.…”

Section: Methodsmentioning

confidence: 99%

“…Acquired dMRI data was processed in the Oxford FMRIB Software Library (FSL) [14]), to estimate diffusion tensor imaging (DTI) parameters, such as fractional anisotropy (FA) maps, in subject-specific spaces. Since FA features have been reported in earlier work to be promising in characterizing PTS [15], [16], we focused on extracting and utilizing FA features in this work.…”

Section: A Data Collection and Preparationmentioning

confidence: 99%

“…However, to our knowledge, no earlier work investigated the classification of LPTS from an incomplete multimodal dataset, proposed suitable novel fusion algorithms, and searched for modalityspecific potential biomarkers. Aside, the current work builds on our earlier preliminary works in preprocessing and analyzing the data (Akbar et al, 2020(Akbar et al, , 2021Faghihpirayesh et al, 2021) from the ongoing multi-center EpiBioS4Rx cohort.…”

Section: Related Workmentioning

confidence: 99%

See 4 more Smart Citations

Advancing Post-Traumatic Seizure Classification and Biomarker Identification: Information Decomposition Based Multimodal Fusion and Explainable Machine Learning with Missing Neuroimaging Data

Akbar

Ruf

Singh

et al. 2022

Preprint

Self Cite

View full text Add to dashboard Cite

Late post-traumatic seizure (LPTS) is a complication of traumatic brain injury (TBI), which can lead to a potentially lifelong condition of post-traumatic epilepsy (PTE). Currently, the patho-mechanism that induces epileptogenesis in TBI subjects is unclear. As such, the epilepsy community strives to identify which TBI subjects will develop epilepsy and find potential biomarkers. To that end, this study collects longitudinal multimodal data from TBI subjects at multiple participating institutes. A supervised, binary classification task is formed with data from the LPTS versus no LPTS subjects. Missing modalities in certain subjects is handled in two ways. First, we extend a graphical model based Bayesian estimator to directly classify subjects with missing modality, and second, we investigate standard imputation techniques. The multimodal information is then combined, following several fusion and dimensionality reduction techniques found in literature, and eventually fitted to a kernelor a tree-based classifier. For this fusion, we propose two new algorithms: recursive elimination of correlated components (RECC) which filters information based on correlation, and information decomposition and selective fusion (IDSF) which meaningfully recombines information from decomposed multimodal features. Based on the cross-validated area under the curve (AUC) score, we find the proposed IDSF algorithm provides the best performance. Finally, following statistical analyses of the frequently selected features, we recommend alterations in inferior temporal gyrus as a potential biomarker.

show abstract