Computational Intelligence最新文献

Lesion detection and segmentation are essential yet complex tasks in medical image analysis due to the substantial variability in lesion shape, size, contrast, and anatomical location across different organs. Existing deep learning methods often lack adaptability, as they are typically designed for specific organs or imaging modalities, leading to limited generalization when applied to diverse datasets. To address this limitation, this study introduces a unified and generalizable framework capable of accurate multi-organ lesion detection, localization, and segmentation across heterogeneous medical imaging data. The proposed U-VQVAE-CTLesionNet integrates a U-Net–based encoder–decoder architecture for spatial feature extraction with a Vector Quantized Variational Autoencoder (VQVAE) module that discretizes latent features through a learnable codebook, enabling the network to capture intricate texture and intensity variations while preserving structural consistency. A Bounding Box Regression (BBR) component is incorporated for lesion localization, followed by a GrabCut-based refinement step that iteratively adjusts lesion boundaries using Gaussian Mixture Model estimation and graph-cut optimization. The framework is further supported by a comprehensive preprocessing pipeline involving intensity normalization, Hounsfield Unit windowing, and affine transformations to standardize image quality and enhance model robustness across modalities. Comprehensive experiments conducted on multiple publicly available and locally curated datasets encompassing lung and kidney lesions validated the accuracy and stability of the proposed approach. For lung CT detection, the model achieved 98.8% accuracy, 98.0% precision, 97.03% recall, and a 97.51% F1-score, while kidney CT detection attained 99.1% accuracy, 99.0% precision, 98.8% recall, and a 98.9% F1-score. Segmentation performance yielded Dice coefficients of 96.5% for lung and 97.8% for kidney, with corresponding IoU values of 93.2% and 95.1%, and Hausdorff Distances of 2.8 mm for lung and 2.3 mm for kidney, respectively. Ablation studies further confirmed that the inclusion of preprocessing, quantization, BBR, and GrabCut modules improved segmentation accuracy by approximately 2%–3% compared to configurations without these components. These results demonstrate that U-VQVAE-CTLesionNet provides a robust, organ-agnostic framework for precise lesion analysis and establishes a solid foundation for future expert-assisted clinical integration.

This study develops an AI-enhanced smart classroom framework to overcome the limitations of traditional English translation instruction, particularly in addressing the theory-practice gap. We designed a three-phase intelligent teaching system incorporating Hidden Markov Models (HMM) for: (1) adaptive pre-class preparation, (2) immersive virtual translation scenarios, and (3) automated post-class assessment. A 6-week controlled experiment (N = 100) compared this approach with traditional instruction using quantitative metrics including engagement levels, translation accuracy, proficiency rates, and satisfaction surveys. The experimental group showed statistically significant improvements (p < 0.05): 12.7% higher engagement (d = 1.21), 8.3% better culture-specific translation accuracy, 6.9% faster proficiency attainment, and 89.2% satisfaction rate (vs. 82.1% control). HMM analysis effectively tracked learning progression and identified competency gaps. The study demonstrates HMM's effectiveness for modeling translation competence development and validates AI-enhanced instruction as a viable solution for translation education. The implemented framework offers a replicable model for intelligent language learning systems.

In traditional control methods, Series Elastic Actuator (SEA) joint manipulators are limited by their hardware and can only perform tasks with low stiffness requirements. To address this issue, we propose a stiffness adjustment control strategy for SEA manipulators based on Dynamic Systems (DS) with an adaptive stiffness control strategy, which could achieve higher manipulator end stiffness comprehensively considering the overall orientation of the manipulator and control gains. By incorporating a rotation matrix with the Dynamic Movement Primitives (DMPs) method, we enhanced the generalization capability of DS in complex tasks. Compared to traditional DS-based control strategies, the control strategy proposed in this paper has achieved better control effects in the experiments on the SEA manipulator, and adaptively adjusts the interaction stiffness under different postures to achieve the task goals. Furthermore, experiments on the mobile manipulator have also verified the universality of the control strategy proposed in this paper.

The exponential growth of online multimedia content across various platforms has created an urgent need for robust assistive technologies to manage the overwhelming volume of information. Consequently, considerable efforts have been dedicated to developing sophisticated multimedia recommendation systems, with Neural Collaborative Filtering (NCF) emerging as a prevalent methodology. However, the conventional NCF method exhibits significant limitations, particularly in integrating contextual data and effectively handling sparse and imbalanced datasets. To address these limitations, this research introduces the Contextual Neural Collaborative Filtering (C-NCF) method, which enhances the NCF framework by incorporating contextual data to enrich the learning process of user-item interactions. The primary objective of this method is to improve rating prediction accuracy, a crucial factor in generating more effective recommendations. A key innovation of the C-NCF method lies in its interaction mechanism, where user ratings of items are evaluated under diverse contextual conditions, assigning varying importance to each contextual factor. Extensive testing on three real-world datasets demonstrated that the C-NCF method outperforms existing advanced techniques. Empirical findings demonstrate that the C-NCF method achieved an average error reduction of 36.43% in Mean Absolute Error and 36.60% in Root Mean Squared Error compared to traditional collaborative filtering, matrix factorization, and context-aware models, significantly enhancing recommendation quality. These insights open promising avenues for further exploration in the field of context-aware recommender systems.

The influence of the COVID-19 pandemic on respiratory health is noteworthy, which tends to exacerbate continuous challenges in global healthcare systems. Chest X-ray (CXR) analysis is considered a significant tool for evaluating the impact of COVID-19 on the lungs. Anyhow, the prevailing analysis approaches frequently seem to be highly complex or cannot activate diagnosis and classify the disease precisely. To overcome these challenges, a novel multimodal deep learning framework that integrates advanced feature extraction, optimization, and classification approaches is proposed for enhanced detection of COVID-19 and severity assessment. The proposed framework introduces a novel hybrid spatial attention assisted extended version of the LSTM and GRU model (HAS-E-LSTM-GRU) that integrates long and short-term temporal feature extraction and attention approach. The prolonged framework maximizes the number of hidden units, whereas the learning and complex patterns obtained from CXR images represent a key novelty of the proposed model. The extracted features are enhanced further using Fourier, Wavelet, Convolutional, Gabor, and Cosine transforms. It is pursued by a feature selection approach using Grey Wolf Optimizer (GWO) to determine the most significant features with higher interclass variance. For the detection of disease, an ensemble classifier integrating Naïve Bayes (NB), k Nearest Neighbors (KNN), Deep Forest (DF), and Support Vector Machine (SVM) is used to determine the COVID-19 presence or absence. Finally, the disease severity is evaluated using a self-attention-based 1D convolutional neural network (CNN) that classifies the diverse cases into mild, moderate, and severe. In association with the standard deep learning methodologies, the proposed architecture obtains a detection accuracy of 98.51% which illustrates the capability in the case of real-time clinical applications.

Automated detection of mental conditions from text is increasingly important due to the rise of digital communication and limited access to clinicians, but the methods often suffer from insufficient explainability. In this study, we propose a novel framework to utilize linguistic dimensions to explain the attribution scores (Integrated Gradients) of tokens in a deep neural network (HerBERT) trained to detect autism and schizophrenia from textual data. In our approach, the scores are mapped to syntactic and psycholinguistic variables, which are then used to train explainable models. The feature importance derived from these models is thereby linked to linguistic dimensions, enhancing interpretability. Our results emphasize the role of syntactic clues, such as grammatical gender and case, in language processing. We also demonstrate that the deep model's attribution scores, when mapped to linguistic features, correlate with selected clinical tests. We propose that our framework can enhance the analysis and interpretation of deep neural networks used for detecting psychiatric disorders. Our approach advances explainable artificial intelligence in mental health by providing researchers with deeper insights into the linguistic cues driving deep neural model predictions.

Existing conversational AI models lack effective multimodal fusion, leading to poor contextual understanding and factual inconsistencies. This research proposes a novel multimodal deep learning framework that integrates vision (CNNs and Vision Transformers), text (Transformers like BERT and GPT-4), and speech (RNNs and Self-Supervised Learning) using an attention-based fusion mechanism. The model enhances sarcasm detection and sentiment analysis through contrastive learning, while graph neural networks (GNNs) ensure knowledge-augmented reasoning, reducing hallucinations in AI-generated responses. Reinforcement learning optimizes dialogue generation for more natural and contextually accurate interactions. Experimental evaluations on MELD, CMU-MOSEI, and AVSD datasets demonstrate a 9%–12% improvement in classification accuracy, 15% higher sarcasm detection efficiency, and a 23% reduction in factual inconsistencies compared to state-of-the-art models like GPT-4, CLIP, and Flamingo. Extensive benchmarking using precision-recall analysis and response fluency metrics validates the robustness of the proposed framework in multimodal sentiment classification and contextual alignment. This research lays the foundation for next-generation multimodal conversational AI, with potential applications in real-world domains such as virtual assistants, healthcare diagnostics, and financial intelligence systems.

Botnet attacks pose a substantial risk to Internet of Things (IoT) settings by using interconnected nodes for malicious intentions, including infiltrating networks or initiating Distributed Denial of Service (DDoS) attacks. To overwhelm systems and cause service interruptions and security breaches, these attacks take advantage of compromised IoT devices. Given the growing prevalence of IoT devices and their associated vulnerabilities, an effective botnet attack detection model is crucial for safeguarding network integrity. To address this critical issue, a new botnet attack detection approach based on the Improved Bi-LSTM-LinkNet model (IBLLNet) is proposed, which is specifically implemented for IoT settings. Preprocessing, feature extraction, feature selection, and detection are the four main stages of this methodology. To solve class imbalance, the preprocessing phase makes use of the Enhanced Synthetic Minority Over-sampling Technique (ESMOTE) method, which produces more representative synthetic data than the traditional SMOTE algorithm. During feature extraction, the proposed approach captures diverse network behaviors through raw, statistical, and entropy-based features. This hybrid approach utilizes Correlation-based Feature Selection (CFS), Recursive Feature Elimination (RFE), and Ridge (L2 regularization) methods, and incorporates an improved selection process where feature weights are assigned based on their predicted significance. In the detection phase, the IBLLNet model integrates an Improved Bidirectional Long Short-Term Memory (IBi-LSTM) and LinkNet models. Here, the IBi-LSTM model incorporates advanced layers like Bi-LSTM, Attentive Context Normalization (ACN), and Artificial Neural Network (ANN) layers to increase detection efficiency and accuracy. The effectiveness of this strategy is confirmed by thorough tests against a number of performance metrics and comparisons with current strategies, indicating its resilience in addressing security risks associated with botnet assaults in Internet of Things environments.

Requirements engineering is one of the most crucial parts of the lifecycle of software engineering. Many programs fail annually due to deficiencies in requirements engineering. Requirements engineering documents are written in natural languages, which can lead to ambiguities. The presence of ambiguity in natural language causes misunderstandings. Accurate and timely identification of these requirements is vital for the development process. However, manual classification is time-consuming and necessitates automation. Today, with the rapid advancement of technology, machine learning and deep learning are being used to detect these ambiguities in requirement specification documents. The BERT word embedding technique and the Bi-LSTM algorithm were used in this research. We have used meta-heuristic algorithms to choose the best value of hyperparameters of our deep learning algorithm. The publicly available Fault-Prone SRS dataset was utilized to train the models. This dataset was also used to evaluate the performance of the proposed algorithm in terms of F1-score, accuracy, and other statistical metrics. The BERT-BiLSTM model outperformed other models in classifying and detecting ambiguities in requirement specification documents, achieving an F1-score and higher than 81% accuracy.