Automated interictal epileptiform discharge (IED) detection has been widely studied, with machine learning methods at the forefront in recent years. As computational resources become more accessible, researchers have applied deep learning (DL) to IED detection with promising results. This systematic review aims to provide an overview of the current DL approaches to automated IED detection from scalp electroencephalography (EEG) and establish recommendations for the clinical research community. We conduct a systematic review according to the PRISMA guidelines. We searched for studies published between 2012 and 2022 implementing DL for automating IED detection from scalp EEG in major medical and engineering databases. We highlight trends and formulate recommendations for the research community by analyzing various aspects: data properties, preprocessing methods, DL architectures, evaluation metrics and results, and reproducibility. The search yielded 66 studies, and 23 met our inclusion criteria. There were two main DL networks, convolutional neural networks in 14 studies and long short-term memory networks in three studies. A hybrid approach combining a hidden Markov model with an autoencoder was employed in one study. Graph convolutional network was seen in one study, which considered a montage as a graph. All DL models involved supervised learning. The median number of layers was 9 (IQR: 5–21). The median number of IEDs was 11 631 (IQR: 2663–16 402). Only six studies acquired data from multiple clinical centers. AUC was the most reported metric (median: 0.94; IQR: 0.94–0.96). The application of DL to IED detection is still limited and lacks standardization in data collection, multi-center testing, and reporting of clinically relevant metrics (i.e. F1, AUCPR, and false-positive/minute). However, the performance is promising, suggesting that DL might be a helpful approach. Further testing on multiple datasets from different clinical centers is required to confirm the generalizability of these methods.

Background and Objectives: To examine the preferences and user experiences of people with epilepsy and caregivers regarding automated wearable seizure detection devices. Methods: We performed a mixed-methods systematic review. We searched electronic databases for original peer-reviewed publications between January 1, 2000, and May 26, 2021. Key search terms included “epilepsy”, “seizure”, “wearable”, and “non-invasive”. We performed a descriptive and a qualitative thematic analysis of the studies included according to the technology acceptance model. Full texts of the discussion sections were further analyzed to identify word frequency and word mapping. Results: Twenty-two observational studies were identified. Collectively, they comprised responses from 3299 participants including patients with epilepsy, caregivers and healthcare workers. Sixteen studies examined user preferences, five examined user experiences, and one examined both experiences and preferences. Important preferences for wearables included improving care, cost, accuracy, and design. Patients desired real-time detection with a latency of ≤15 minutes from seizure occurrence, along with high sensitivity (≥90%) and low false-alarm rates. Device related costs were a major factor for device acceptance, where device costs of <$300 USD and a monthly subscription fee of <$20 USD were preferred. Despite being a major driver of wearable-based technologies, sudden unexpected death in epilepsy (SUDEP) was rarely discussed. Among studies evaluating user experiences, there was a greater acceptance towards wristwatches. Thematic coding analysis showed that attitudes towards device use, and perceived usefulness were reported consistently. Word mapping identified ‘specificity’, ‘cost’, and ‘battery’ as key single terms, and ‘battery life’, ‘insurance coverage’, ‘prediction/detection quality’, and the effect of devices on ‘daily life’ as key bigrams. Discussion: User acceptance of wearable technology for seizure detection was strongly influenced by accuracy, design, comfort, and cost. Our findings emphasise the need for standardised and validated tools to comprehensively examine preferences and user experiences of wearable devices in this population, using the themes identified in this study. Greater efforts to incorporate perspectives and user experiences in developing wearables for seizure detection, particularly in community-based settings are needed. PROSPERO Registration: CRD42020193565.

Continuous fetal movement monitoring in late pregnancy may improve fetal wellbeing and pregnancy outcomes. While fetal movements can be visualized with ultrasound, it is intermittent and limited to clinical settings. Inertial measurement units may enable at-home fetal monitoring but usually require a large-footprint, multisensor design. Here, we report smart, compact wearable pressure-strain combo sensors continuously monitoring fetal movements through maternal abdominal skin motions. In the 2D and 3D artificial abdomen systems, our octagonal-shaped gold nanowire–based strain sensor served as an isotropic sensor, enabling omnidirectional simulated “kicking load” detection within an area of ~77 (2D) and ~217 cm ² (3D), while an interdigitated electrode–based pressure sensor showed highly sensitive localized load detection. Building upon these findings, we designed compact pressure- and strain-sensing integrated Band-Aids and tested on 59 pregnant women. We developed machine learning models to distinguish fetal from nonfetal movements with >90% accuracy in ultrasound-based validation studies. This AI-powered, Band-Aid–like sensing system offers potential as a compact, comfortable, and accurate continuous out-of-hospital fetal movement monitoring technology.

Epilepsy is a common neurologic disorder. Although antiseizure medications (ASMs) are the first-line treatment, identifying the most effective ASM for each individual remains a trial-and-error process. Genetic variation may influence treatment response. We aimed to develop and validate a multimodal deep learning model that integrates clinical and genomic features to predict response to the initial ASM in people with newly diagnosed epilepsy. We used data from individuals with newly diagnosed epilepsy in Australia as the development cohort and participants from the Human Epilepsy Project 1 (recruited in the United States, Europe, and Australia) as the external validation cohort. All participants initiated ASM treatment and were followed prospectively for at least 1 year. We included 16 clinical factors and constructed 4 genomic feature types related to epilepsy and ASM pharmacogenomics, with and without functional impact annotations. We evaluated various machine learning architectures and multimodal fusion strategies to predict seizure freedom while taking the initial ASM at 1 year. In the development cohort (n = 286, median age 39 years, 47.2% seizure free), combining clinical and genomic features in our proposed multimodal deep learning model improved predictive performance. The highest area under the receiver operating characteristic curve (AUC) of 0.74 (95% CI 0.70-0.78) was achieved using clinical factors and genomic variants affecting transcription factor binding, significantly outperforming the clinical-only model (AUC 0.67, 95% CI 0.62-0.72; p < 0.05). In the external validation cohort (n = 219, median age 31 years, 20.5% seizure free), the same feature combination achieved an AUC of 0.69 (95% CI 0.67-0.71), higher than the clinical-only model (AUC 0.62, 95% CI 0.60-0.64; p < 0.05). Applying this model to the development cohort, if all participants took the highest ranked ASMs, the mean predicted seizure-free probability would be 68.05% (95% CI 65.79%-70.35%) compared with the observed seizure-free rate of 47.2% (95% CI 41.3%-53.2%). Integrating genomic data with clinical features enhances the ability of deep learning models in predicting ASM response in newly diagnosed epilepsy. This approach may support personalized treatment selection and improve clinical outcomes.

Accurate prediction of antiseizure medication (ASM) outcomes is crucial for optimising epilepsy treatment. We propose a multi-modal deep learning framework that integrates electroencephalography (EEG), magnetic resonance imaging (MRI), clinical factors, and molecular drug features to enhance ASM outcome prediction. Our approach includes EEG Q-Net, a pre-trained quantisation model capturing finegrained temporal EEG patterns, and MRI embedding from BiomedCLIP. To capture chemical similarities and interactions for the ASMs, we employed a pre-trained model trained on thousands of medications to generate embedding for SMILES (Simplified Molecular Input Line Entry System). Our findings underscore the benefits of multimodal integration for personalised epilepsy management, as our fusion model achieved an AUC of 0.75, an improvement over the best unimodal model (0.71).

Epilepsy is a common neurological disorder, with approximately one-third of the affected population developing drug-resistant epilepsy despite the exponentially increasing number of antiseizure medications (ASMs) that are available. Current ASM selection remains largely empirical, lacking reliable biomarkers to predict treatment response. Pharmacogenomics is an important factor underlying drug response. The development of models designed to predict drug response in epilepsy that include genomic information is at an early stage. Advances in artificial intelligence (AI) and deep learning (DL) offer new opportunities for multimodality modelling and, in particular, models that can reason over both a patient’s genomic risk factors and their clinical features (e.g. epilepsy type, magnetic resonance imaging [MRI]/electroencephalography [EEG] findings). More broadly, new technologies are driving a paradigm shift towards agentic AI, whereby decision-support agents will coordinate medical information and propose actions. These advances have implications for individualised ASM prescribing in epilepsy. For example, rather than modelling thousands of genetic variants, a DL model can operate with compressed representations of data obtained from AI foundation models that have been trained on huge atlases to learn representations of general biology, enhancing capacity for small epilepsy cohorts. In this review, we summarise how emerging technologies are being harnessed and highlight strategies for building capacity. We specifically focus on clinical, pharmacological, and genomic data, discussing the ways in which drug information can be formatted to take advantage of DL model capabilities. We describe AI foundation models and provide examples of how they can be utilised in epilepsy research. We conclude with a summary of ongoing challenges, including model evaluation, data availability, ethical considerations, and barriers to clinical translation.