Attention-based deep convolutional neural network for classification of generalized and focal epileptic seizures

Epilepsy affects over 50 million people globally. Electroencephalography is critical for epilepsy diagnosis, but manual seizure classification is time-consuming and requires extensive expertise. This paper presents an automated multi-class seizure classification model using EEG signals from the Temple University Hospital Seizure Corpus ver. 1.5.2. 11 features including time-based correlation, time-based eigenvalues, power spectral density, frequency-based correlation, frequency-based eigenvalues, sample entropy, spectral entropy, logarithmic sum, standard deviation, absolute mean, and ratio of Daubechies D4 wavelet transformed coefficients were extracted from 10-second sliding windows across channels. The model combines multi-head self-attention mechanism with a deep convolutional neural network (CNN) to classify seven subtypes of generalized and focal epileptic seizures. The model achieved 0.921 weighted accuracy and 0.902 weighted F1 score in classifying focal onset non-motor, generalized onset non-motor, simple partial, complex partial, absence, tonic, and tonic-clonic seizures. In comparison, a CNN model without multi-head attention achieved 0.767 weighted accuracy. Ablation studies were conducted to validate the importance of transformer encoders and attention. The promising classification results demonstrate the potential of deep learning for handling EEG complexity and improving epilepsy diagnosis. This seizure classification model could enable timely interventions when translated into clinical practice.

Optimal designing of grid-connected microgrid systems for residential and commercial applications in Pakistan

Conventional energy sources (CESs) are currently serving most of the global energy demands, but they will be substantially depleted as moving towards the end of this decade. The generation of electricity from such sources is causing the emission of greenhouse gases that is resulting in deleterious effect on the environment along with changing climatic and energy patterns of the planet. Therefore, the world is heading toward decentralization, and microgrids are playing a key role in this process. The advantages of renewables, which are acknowledged globally as benign, eco-friendly, economical, and inexhaustible resources available worldwide, are to credit for such a massive surge in the utilization of renewable resources in microgrid technology. Despite having an enormous renewable energy potential, Pakistan spends a sizable portion of its budget on energy imports of coal, oil, and liquefied natural gas, however, with good planning, current energy crises might be eliminated or at least mitigated to a greater extent, assuring energy security, economic prosperity, and lower carbon emissions inside the country. This study considers the optimal component planning in a grid-connected microgrid with five objectives to achieve that are to reduce the cost of energy, increase the renewable share, cut greenhouse gas emissions, enhance the reliability of power supply and to make electricity generation sustainable in the long run for the country. Different solar PV capacities are tested against the energy cost, renewable share and emission of greenhouse gases in order to attain the trade-off. The cost of energy is minimized by 92.47%, renewable share rises to 85%, and CO₂ emissions are decreased by 48% for residential application. In the case of commercial application, however, the cost of energy is lowered by 48.52%, the renewable energy share rises to 71.1%, and CO₂ emissions are reduced by 61% through incorporating solar PV into the current power system.

Seizure forecasting using single robust linear feature as correlation vector of seizure-like events in brain slices preparation in vitro

OBJECTIVE

Epilepsy is a neurological disorder affecting 50 million individuals globally. Modern research has inspected the likelihood of forecasting epileptic seizures. Algorithmic investigations are giving promising results for seizure prediction. Though mostly seizure prediction algorithm uses pre-ictal (prodromal symptoms) events for prediction. On the contrary, prodromal symptoms may not necessarily be present in every patient or subject. This paper focuses on seizure forecasting regardless of the presence of pre-ictal (prodromal symptoms) using the single robust feature with maximum accuracy. Method: We evaluated datasets having 4-aminopydine induced seizure-like events rat's hippocampa slices and cortical tissue from pharmacoresistant epilepsy patients. The proposed methodology applies the Discrete Wavelet Transform (DWT) at levels 1-5 utilizing 'Daubechies-4'. Linear Discriminant classifier (LDC), Quadratic Discriminant Classifier (QDC) and Support Vector Machine (SVM) were used to classify each signal using eight discriminative features. Results: Classifier performance was assessed by parameters like true detections (TD), false detection (FD), accuracy (ACC), sensitivity (SEN), specificity (SPF), and positive predicted value (PPC), negative predicted value (NPV). Highest classification feature was selected as a seizure forecasting correlation vector and decision rule was formulated for seizure forecasting. Correlation vector served as a forecaster for current EEG activity. Proposed decision rule forecasted ongoing signal activity towards possible seizure condition true or false. The suggested framework revealed forecasting of ictal events at 10 seconds before the actual seizure. Conclusion: It is worth mentioning that the proposed study utilized a single linear feature to predict seizures precisely. Moreover, utilization of single feature encouraged in subsiding system complexity, processing delays, and system latency.