In this article, we propose a circuit to imitate the behavior of a Reward-Modulated spike-timing-dependent plasticity synapse. When two neurons in adjacent layers produce spikes, each spike modifies the thickness in the shared synapse. As a result, the synapse’s ability to conduct impulses is controlled, leading to an unsupervised learning rule. By introducing a reward signal, reinforcement learning is enabled by redirecting the growth and shrinkage of synapses based on signal feedback from the environment. The proposed synapse manages the convolution of the emitted spike signals to promote either the strengthening or weakening of the synapse, represented as the resistance value of a memristor device. As memristors have a conductance range that may differ from the available current input range of typical CMOS neuron designs, the synapse circuit can be adjusted to regulate the spike’s amplitude current to comply with the neuron. The circuit described in this work allows for the implementation of fully interconnected layers of neuron analog circuits. This is achieved by having each synapse reconform the spike signal, thus removing the burden of providing enough power from the neurons to each memristor. The synapse circuit was tested using a CMOS analog neuron described in the literature. Additionally, the article provides insight into how to properly describe the hysteresis behavior of the memristor in Verilog-A code. The testing and learning capabilities of the synapse circuit are demonstrated in simulation using the Skywater-130 nm process. The article’s main goal is to provide the basic building blocks for deep neural networks relying on spiking neurons and memristors as the basic processing elements to handle spike generation, propagation, and synaptic plasticity.

Early diagnosis of cognitive decline is vital for timely treatment of mild cognitive impairment (MCI) and Alzheimer’s disease (AD), yet standard clinical assessments often miss subtle longitudinal language changes. We propose a hierarchical hybrid intelligence framework integrating long-context language modeling, temporal progression, semantic graph reasoning, psycholinguistic biomarkers, and contrastive progression learning to classify patient states (Normal, MCI, AD) from longitudinal electronic health record (EHR) notes. The model was trained on 4500 patients and 68,000 clinical notes from Medical Information Mart for Intensive Care III (MIMIC-III) and externally validated on the Medical Information Mart for Intensive Care IV (MIMIC-IV) clinical notes dataset (5200 patients, 72,000 notes). Inputs combined Biomedical and Clinical Bidirectional Encoder Representations from Transformers (BioClinicalBERT) embeddings, Bidirectional Long Short-Term Memory (Bi-LSTM) temporal encodings, Graph Sample and Aggregate (GraphSAGE)-based Unified Medical Language System (UMLS) concept graphs, and psycholinguistic vectors (lexical diversity, grammatical complexity, discourse coherence). On the MIMIC-III hold-out set, the model achieved 99.999% accuracy, a macro F1-score of 0.999, a Receiver Operating Characteristic Area Under the Curve (ROC AUC) of 0.999, and a temporal stability variance of 0.0008. Monte Carlo cross-validation (10,000 folds) yielded $99.997\pm 0.003$% accuracy and $0.999\pm 0.001$ macro F1. Feature ablation confirmed distinct gains from temporal, semantic, and psycholinguistic modules, improving performance by 1.1% over text-only baselines. Cross-cohort zero-shot testing on MIMIC-IV showed strong generalization with minimal decline in macro F1 and balanced accuracy. Explainability analyses, such as SHapley Additive exPlanations (SHAP) token/concept attribution, attention maps, counterfactual perturbations, and psycholinguistic importance, revealed clinically interpretable markers, such as pronoun overuse, reduced lexical diversity, and syntactic simplification, as predictors of decline. Our framework supports scalable, non-invasive early screening in a variety of healthcare settings by providing longitudinally stable predictions.

This study presents FLIGENT-Net, a multi-domain signal representation framework for gastric histopathological image classification that treats tissue images as complex two-dimensional signals whose diagnostic content is distributed across spatial, spectral, relational, and contextual domains. Using the GasHisSDB dataset, FLIGENT-Net investigates spatial representations extracted through a convolutional neural network (CNN), frequency-domain information obtained using the Fast Fourier Transform (FFT), graph-based relational modeling through a Graph Convolutional Network (GCN) acting as a graph-signal-processing operator, and global contextual modeling through a Vision Transformer (ViT). Rather than evaluating the framework solely according to classification accuracy, a systematic characterization was performed using individual and combined configurations, computational cost, training dynamics, interpretability, Pearson and Spearman correlation of importance maps, and Pixel Deletion analysis. FLIGENT-Net achieved 96.98% Accuracy, 95.45% Precision, 96.95% Recall, 96.20% F1-Score, and 0.9959 AUC-ROC. Notably, the simpler CNN+GCN configuration achieved a higher Accuracy of 97.76%, while CNN+GCN+FFT achieved 97.75%; this outcome shows that combining all four representation domains does not monotonically improve classification accuracy, reinforcing the characterization-oriented contribution of this study over a purely performance-driven one. The importance-map analysis revealed distinct patterns across representations, with low global Pearson and Spearman correlations. Pixel Deletion yielded a positive ∆AUC of 0.1450 for FLIGENT-Net. Overall, the results characterize distinct classification, computational, interpretability, and faithfulness profiles across multi-domain signal representations, offering a computational methods case study in spectral and graph-based image-signal analysis for biomedical applications.

This study characterizes spatial, spectral, relational, and contextual representations for binary gastric histopathological image classification using FLIGENT-Net. The framework combines an ImageNet-pretrained convolutional backbone, patch-level fast Fourier transform (FFT) descriptors, graph convolution over a cosine-similarity graph, and a Transformer Encoder. Experiments use the 160-by-160 GasHisSDB subset, a persisted stratified 60%/20%/20% partition, and one training run per configuration with seed 42. In the nine completed configurations of the revised evaluation, FLIGENT-Net achieved 97.31% accuracy, 96.57% precision, 96.61% recall, 96.59% F1-score for the Abnormal class, and 0.9964 AUC-ROC. An additional native-grid FFT control yielded 97.43% accuracy and 96.74% F1 under an alternative spectral protocol that preserved the 160-by-160 sampling grid while changing the local FFT window sizes and spectral discretization. The conventional CNN baseline (ResNet18) obtained the highest accuracy and F1-score among these completed runs, at 98.51% and 98.13%, respectively. These results do not establish a predictive advantage for the complete multi-domain architecture; rather, they characterize how the added representation domains affect predictive behavior, graph topology, computational cost, and attribution response under a common protocol. The study contributes a reproducible integration of the four representation domains and a controlled empirical characterization of their combinations. Transformer attention visualizations describe information allocation among contextualized nodes, while a common input-gradient deletion protocol examines attribution behavior. Together, these analyses identify the behavior and limitations of the implemented fusion strategy and provide a basis for its further development. Performance estimates are conditional on the available patch-level split and should not be interpreted as patient-, slide-, or source-image-independent generalization; run-to-run variability is also not estimated because one fixed seed is used per configuration.