The success of arrhythmia classification tasks with Machine Learning (ML) algorithms is based on the handcrafting extraction of features from Electrocardiography (ECG) signals. However, feature extraction is a time-consuming trial-and-error approach. Deep Neural Network (DNN) algorithms bypass the process of handcrafting feature extraction since the algorithm extracts the features automatically in their hidden layers. However, it is important to have access to a balanced dataset for algorithm training. In this exploratory research study, we will compare the evaluation metrics among Convolutional Neural Networks (1D-CNN) and Support Vector Machines (SVM) using a dataset based on the merged public ECG signals database TNMG and CINC17 databases. Results: Both algorithms showed good performance using the new, merged ECG database. For evaluation metrics, the 1D-CNN algorithm has a precision of 93.04%, an accuracy of 93.07%, a recall of 93.20%, and an F1-score of 93.05%. The SVM classifier ($\lambda$ = 10, C = 10 × 10⁹) achieved the best classification metrics with two combined, handcrafted feature extraction methods: Wavelet transforms and R-peak Interval features, which achieved an overall precision of 89.04%, accuracy of 92.00%, recall of 94.20%, and F1-score of 91.54%. As an unique input feature and SVM ($\lambda =10,C=100$), wavelet transforms achieved precision, accuracy, recall, and F1-score metrics of 86.15%, 85.33%, 81.16%, and 83.58%. Conclusion: Researchers face a challenge in finding a broad dataset to evaluate ML models. One way to solve this problem, especially for deep learning models, is to combine several public datasets to increase the amount of data. The SVM and 1D-CNN algorithms showed positive results with the merge of databases, showing similar F1-score, precision, and recall during arrhythmia classification. Despite the favorable results for both of them, it should be considered that in the SVM, feature selection is a time-consuming trial-and-error process; meanwhile, CNN algorithms can reduce the workload significantly. The disadvantage of CNN algorithms is that it has a higher computational processing cost; moreover, in the absence of access to powerful computational processing, the SVM can be a reliable solution.

The development of mobile health for epilepsy has grown in the last years, bringing new applications (apps) to the market and improving already existing ones. In this systematic review, we analyse the scope of mobile apps for seizure detection and epilepsy self-management, with two research questions in mind: what are the characteristics of current solutions and do they meet users’ requirements? What should be considered when designing mobile health for epilepsy? We used PRISMA methodology to search within App Store and Google Play Store from February to April of 2021, reaching 55 potential apps. A more thorough analysis regarding particular features was performed on 26 of those apps. The content of these apps was evaluated in five categories, regarding if there was personalisable content; features related to medication management; what aspects of seizure log were present; what type of communication prevailed; and if there was any content related to seizure alarm or seizure action plans. Moreover, the 26 apps were evaluated through using MARS by six raters, including two neurologists. The analysis of MARS categories was performed for the top and bottom apps, to understand the core differences. Overall, the lowest MARS scores were related to engagement and information, which play a big part in long-term use, and previous studies raised the concern of assuring continuous use, especially in younger audiences. With that in mind, we identified conceptual improvement points, which were divided in three main topics: customisation, simplicity and healthcare connection. Moreover, we summarised some ideas to improve m-health apps catered around long-term adherence. We hope this work contributes to a better understanding of the current scope in mobile epilepsy management, endorsing healthcare professionals and developers to provide off-the-shelf solutions that engage patients and allows them to better manage their condition.

This study aims to review the proposed methodologies and reported performances of automated algorithms for seizure forecast. A systematic review was conducted on studies reported up to May 10, 2024. Four databases and registers were searched, and studies were included when they proposed an original algorithm for automatic human epileptic seizure forecast that was patient specific, based on intraindividual cyclic distribution of events and/or surrogate measures of the preictal state and provided an evaluation of the performance. Two meta-analyses were performed, one evaluating area under the ROC curve (AUC) and another Brier Skill Score (BSS). Eighteen studies met the eligibility criteria, totaling 43 included algorithms. A total of 419 patients participated in the studies, and 19442 seizures were reported across studies. Of the analyzed algorithms, 23 were eligible for the meta-analysis with AUC and 12 with BSS. The overall mean AUC was 0.71, which was similar between the studies that relied solely on surrogate measures of the preictal state, on cyclic distributions of events, and on a combination of these. BSS was also similar for the three types of input data, with an overall mean BSS of 0.13. This study provides a characterization of the state of the art in seizure forecast algorithms along with their performances, setting a benchmark for future developments. It identified a considerable lack of standardization across study design and evaluation, leading to the proposal of guidelines for the design of seizure forecast solutions.