One outstanding challenge for machine learning in diagnostic biomedical imaging is algorithm interpretability. A key application is the identification of subtle epileptogenic focal cortical dysplasias (FCDs) from structural MRI. FCDs are difficult to visualize on structural MRI but are often amenable to surgical resection. We aimed to develop an open-source, interpretable, surface-based machine-learning algorithm to automatically identify FCDs on heterogeneous structural MRI data from epilepsy surgery centres worldwide. The Multi-centre Epilepsy Lesion Detection (MELD) Project collated and harmonized a retrospective MRI cohort of 1015 participants, 618 patients with focal FCD-related epilepsy and 397 controls, from 22 epilepsy centres worldwide. We created a neural network for FCD detection based on 33 surface-based features. The network was trained and cross-validated on 50% of the total cohort and tested on the remaining 50% as well as on 2 independent test sites. Multidimensional feature analysis and integrated gradient saliencies were used to interrogate network performance. Our pipeline outputs individual patient reports, which identify the location of predicted lesions, alongside their imaging features and relative saliency to the classifier. On a restricted ‘gold-standard’ subcohort of seizure-free patients with FCD type IIB who had T1 and fluid-attenuated inversion recovery MRI data, the MELD FCD surface-based algorithm had a sensitivity of 85%. Across the entire withheld test cohort the sensitivity was 59% and specificity was 54%. After including a border zone around lesions, to account for uncertainty around the borders of manually delineated lesion masks, the sensitivity was 67%. This multicentre, multinational study with open access protocols and code has developed a robust and interpretable machine-learning algorithm for automated detection of focal cortical dysplasias, giving physicians greater confidence in the identification of subtle MRI lesions in individuals with epilepsy.

This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited.

Introduction: The overall combined prevalence of anxiety and depression in patients with epilepsy has been estimated at 20.2 and 22.9%, respectively, and is considered more severe in drug-refractory epilepsy. Patients admitted to epilepsy monitoring units constitute a particular group. Also, patients with psychogenic non-epileptic seizures can reach more than 20% of all admissions. This study aims to characterize these symptoms in a large cohort of patients admitted for evaluation in a tertiary epilepsy center.Materials and Methods: The study was conducted among 493 consecutive patients (age: 38.78 ± 12.7, 57% females) admitted for long-term video EEG from January 2013 to February 2021. Demographic, clinical, and mood disorder patients' data were collected. Anxiety and depression symptoms were assessed through the Hospital Anxiety Depression Scale (HADS-A and HADS-D), the State Trait Anxiety Inventory (STAI), and Beck Depression Inventory (BDI-II). Quality of life was determined using the QOLIE-10. Patients were divided into three groups: patients with epilepsy (n = 395), psychogenic non-epileptic seizures (PNES) (n = 56), and combined (n = 33). A univariate and multivariate regression analysis was performed for variables associated with quality of life.Results: Of 493 patients, 45.0% had structural etiology, and considering epilepsy classification, 43.6% were of temporal lobe origin. In addition, 32.45% of patients had a previous psychiatric history, 49.9% of patients had depressive symptoms in BDI, and 30.9% according to HADS-D; 56.42 and 52.63% of patients presented pathological anxiety scores in STAI-T and STAI-S, respectively; and 44.78% according to HADS-A. PNES and combined groups revealed a higher incidence of pathologic BDI scores (64.29 and 78.79%, p < 0.001) as well as pathologic HADS-A scores (p = 0.001). Anxiety and depression pathologic results are more prevalent in females, HADS-A (females = 50.7%, males = 36.8%; p = 0.0027) and BDI > 13 (females = 56.6%, males = 41.0%; p = 0.0006). QOLIE-10 showed that 71% of the patients had their quality of life affected with significantly higher scores in the combined group than in the epilepsy and PNES groups (p = 0.0015).Conclusions: Subjective anxiety, depression, and reduced quality of life are highly prevalent in patients with refractory epilepsy. These symptoms are more evident when PNES are associated with epilepsy and more severe among female patients. Most of the cases were not previously diagnosed. These factors should be considered in everyday clinical practice, and specific approaches might be adapted depending on the patient's profile.

The principles governing the formation of episodic memories from the continuous stream of sensory stimuli are not fully understood. Theoretical models of the hippocampus propose that the representational format of episodic memories comprise oscillations in the theta frequency band (2-8 Hz) that set the time boundaries in which discrete events are bound encoded in the gamma frequency range (>30 Hz). We investigated this temporal segmentation and binding process by analyzing the intracranial EEG (iEEG) of surgically implanted epileptic patients performing a virtual-navigation task. We found a positive correlation between sensory information density encountered by the subject and hippocampal theta-frequency, suggesting that the human hippocampus normalizes the information content of episodic memories relative to the density of sensory information. This interpretation is further supported by the observation that as a marker of mnemonic encoding, i.e. the amount of persistent gamma events, directly correlates with sensory information density, gamma-frequency power and the phase relation between theta and gamma oscillations remain constant. Using a theoretical model of the hippocampus, we build a model that analogously displays a similar normalization of gamma episodes per theta cycle relative to information density by accounting for the physiological signatures of theta-gamma coding through combining fast and slow inhibitory feedback. We propose that this intrinsic normalization mechanism optimizes the trade-off between the discretization and compression of continuous experience relative to the limited capacity of episodic memory.