Background: In deep learning-based epilepsy prediction and classification, enhancing the extraction of electroencephalogram (EEG) features is crucial for improving model accuracy. Traditional supervised learning methods rely on large, detailed annotated datasets, limiting the feasibility of large-scale training. Recently, self-supervised learning approaches using masking-and-reconstruction strategies have emerged, reducing dependence on labeled data. However, these methods are vulnerable to inherent noise and signal degradation in EEG data, which diminishes feature extraction robustness and overall model performance. Methods: In this study, we proposed a self-supervised learning Transformer network enhanced with Adaptive Frequency-Time Attention (AFTA) for learning robust EEG feature representations from unlabeled data, utilizing a masking-and-reconstruction framework. Specifically, we pretrained the Transformer network using a self-supervised learning approach, and subsequently fine-tuned the pretrained model for downstream tasks like seizure prediction and classification. To mitigate the impact of inherent noise in EEG signals and enhance feature extraction capabilities, we incorporated AFTA into the Transformer architecture. AFTA incorporates an Adaptive Frequency Filtering Module (AFFM) to perform adaptive global and local filtering in the frequency domain. This module was then integrated with temporal attention mechanisms, enhancing the model’s self-supervised learning capabilities. Result: Our method achieved exceptional performance in EEG analysis tasks. Our method consistently outperformed state-of-the-art approaches across TUSZ, TUAB, and TUEV datasets, achieving the highest AUROC ($0.891$), balanced accuracy ($0.8002$), weighted F1-score ($0.8038$), and Cohen’s kappa ($0.6089$). These results validate its robustness, generalization, and effectiveness in seizure detection and classification tasks on diverse EEG datasets.

Background/Objectives: This study explored the heterogeneous distribution pattern of translocator protein 18kDa (TSPO)-PET/MRI using radioligand [¹⁸F] DPA-714 in temporal lobe epilepsy patients and identified clinical factors influencing imaging outcomes. Methods: The TSPO imaging in individual patient was evaluated with both visual reading and quantitative assessment using an asymmetry index based on cerebellum-normalized standardized uptake values. The association between clinical factors and TSPO imaging outcomes was assessed. Pathological evaluation was conducted in three patients. Results: Twenty-nine TLE patients and ten healthy controls were enrolled. Visual evaluation revealed increased [¹⁸F] DPA-714 uptake in twenty patients as compared to controls, predominantly in a unilateral regional brain, while the remaining nine patients showed visually undetectable uptake of [¹⁸F] DPA-714. Consistently, quantitative analysis revealed that 69% (20/29) patients exhibited at least one brain area with significant asymmetry index, notably in the temporal lobe (85%, 17/20). A high asymmetry index could also be observed in the parietal (13.8%, 4/29) and occipital lobe (17.2%, 5/29). Significant associations were identified between the asymmetry index and seizure frequency (p = 0.045, OR = 7.994), and the interval from last seizure to PET scan (p = 0.033, OR = 6.712). Moreover, we confirmed the pathology in three patients via immunohistochemistry, which underscored the potential of TSPO-PET in detecting minor lesion. Conclusions: TSPO-PET reveals patient-specific and network-level neuroinflammatory heterogeneity in MRI-negative TLE, supporting its potential role as a complementary tool for presurgical evaluation.