Abstracts

Rationale:
Seizure semiology is an important feature in the evaluation of patients with drug resistant epilepsy in order to help lateralise and localise the epileptogenic zone (EZ) for resection. However, the value of initial semiology (iS) and chronological evolution are widely variable. Even strongly localising semiology can localise to disparate regions and the symptomatogenic zone can be incongruent with the EZ._x000D_ In order to surpass these problems with the subjective value of semiology, we created an objective, evidence-based database linking patient semiology to the neuroanatomical EZ to allow more direct predictions.
Method:
We curated the largest patient-level database of 4454 unique patients from 282 studies, yielding 2368 lateralising and 10917 localising datapoints for initial or most prominently reported seizure semiologies. We only included a patient’s semiology according to strict ground-truth criteria: if they had a focal resection and remained seizure free for at least 12 months, or had intracranial stereotactic EEG or stimulation, or concordant neuroimaging and neurophysiology. We also collected data on age to allow filtering out paediatric data. To query this database, we developed a taxonomy of 47 semiological terms, mapped reported categorical brain regions to 55 localising cerebral atlas labels and 5 lateralising possibilities (contra- or ipsi- lateral, dominant or non-dominant hemisphere and bilateral), and developed flexible Bayesian filters for exclusion of patients from studies that preselected patients based on prior knowledge of the EZ, to mitigate the publication bias that favours temporal lobe epilepsy. _x000D_ We integrated the database, taxonomy, mappings, ground truths and Bayesian filters in a Python module for the 3D Slicer program, to create a novel user-friendly and open-source Semiology Visualisation Tool (SVT) which allows 3D-brain visualisations of semiologies and their simultaneous combinations. Our SVT software is available at https://github.com/thenineteen/Semiology-Visualisation-Tool._x000D_ We queried our SVT for a sample of 14 patients who were external to the database and had previously had epilepsy surgery and were subsequently seizure-free, split equally between temporal and extratemporal resections and drawn using a random number generator. We compared the predictive performance of our SVT using different patient-level ground truths and Bayesian filters for iS and set of semiology (SoS), using the scoring method outlined in Table 1.
Results:
The SoS scores are equal or better than iS in all 14 cases for the correct prediction of the EZ. The application of the Bayesian filter improved extratemporal scores (Table 1). A sample output is shown in Figure 1.
Conclusion:
Set of semiology is better than initial semiology for both lateralisation and localisation using our SVT, and Bayesian filtering mitigates publication bias as intended. _x000D_ This data-driven and open-source approach has the potential to be used as clinical decision support for the presurgical evaluation of patients with drug-resistant epilepsy, and as the basis for more complex multimodal models for the determination of the epileptogenic zone.
Funding:
:UCL Wellcome/EPSRC Centre for Interventional and Surgical Sciences (WEISS) (203145Z/16/Z), EPSRC Centre for Doctoral Training in Medical Imaging (EP/L016478/1) and in part independent research commissioned by the Wellcome Trust Health Innovation Challenge Fund (WT106882).
_x000D_
_x000D_