Ear diseases, frequently subject to delayed or erroneous diagnoses, are among the most prevalent challenges encountered by clinicians delivering primary care to children and adolescents. Artificial Intelligence (AI) can aid clinicians in identifying and diagnosing ear diseases via imaging techniques. Therefore, this paper proposes a novel AI-based framework, termed the Fusion-based Multi-Learner Network (FMLN), to enhance the automated diagnosis of ear diseases. In fact, FMLN consists of several steps, (i) Image Preprocessing (IP) and (ii) Disease Classification (DC). Through FMLN two contributions are introduced. The first is a new feature selection methodology that is called Optimized Gray Wolf Optimization (OGWO). OGWO is an advanced version of conventional GWO that incorporates Opposition-Based Learning (OBL) alongside GWO. In fact, OBL assists GWO in concurrently exploring both the current candidate solution and its antithesis, thereby enhancing the likelihood of rapidly identifying superior regions. While the second contribution is a new classification model that is used to perform multi-classification of ear disease using several ML algorithms. The conclusive determination is predicated on weighted majority voting. Experimental results indicated that the proposed FMLN excels in multi-classification tasks for ear diseases. It achieves an overall accuracy of 97.78%, surpassing the latest methodologies.
Data-intensive applications are increasing rapidly and require secure, high-capacity communication technologies. Optical wireless communications based on lasers are a promising field for their high bandwidth, directionality and immunity to electromagnetic interference. But protecting the transmitted data is still a hard problem. This research presents an integrated electro-optical encryption architecture for wireless laser communication systems security. The proposed system uses a coherent laser source based on a Quantum Cascaded Laser (QCL) to provide a stable optical carrier to transmit encrypted signals. The beam combining and collimation subsystem of the transmitter module increases spatial coherence and reduces beam divergence before propagation. The receiver employs wavelength-selective elements and polarisation control devices to decrypt and reconstruct the encoded optical information. The fast steering mirror and tracking subsystem of the receiver's dynamic beam-steering mechanism compensate for gearbox-alignment deviation to ensure signal acquisition. Numerical simulations were used to study the relation between the mirror deflection angles and the spatial position of the received laser spot and to demonstrate the geometric response of the receiver optics. Matrix-based correction increases the stability and the position accuracy of the beam. In addition, the proposed communication scheme uses orbital angular momentum modulation for physical layer encryption. The simulation results indicate that the signals can only be recovered in the right decoding mode so as to guarantee the security of the transmission channel. This way the architecture is compatible with secure high-bandwidth optical communication systems.
Automatic Question Generation (AQG) has emerged as one of the core research problems in educational technology owing to its potential to automate assessments and facilitate the development of adaptive learning platforms. This survey presents a comprehensive study of MCQ generation methods from rule-based and NLP-based approaches to the recent advanced large language models (LLMs) and multimodal systems. Earlier methods were dependent on various linguistic approaches such as Part-of-Speech (POS) tagging, named entity recognition (NER), dependency parsing, Term Frequency-Inverse Document Frequency (TF-IDF), which had certain drawbacks related to fluency, scalability, and distractors. Modern Transformer-based methods have enhanced contextual understanding and question coherence via advanced strategies such as prompt engineering, chain-of-thought reasoning, self-refinement, and Retrieval-Augmented Generation (RAG). Despite recent advances, AQG still struggles with reasoning, distractor quality, and evaluation, while multimodal AQG remains underexplored. Moreover, a separate discussion is presented on multimodal question generation approaches that generate questions based on text, images, and video inputs using vision-language modeling and retrieval. In addition, popularly used methods such as MCQGen, DisGeM, MultiQG-TI, AutoConverter, VQADG, and AGeES are analyzed based on their architecture, datasets, and evaluation measures. Finally, open research issues related to MCQ generation including hallucination, plausibility, and difficulty.
Reliable authentication in vehicular ad-hoc networks remains challenging because safety messages must be verified under strict latency, mobility, and wireless-channel constraints. Conventional cryptographic schemes provide strong security and accountability, but repeated message-level verification may introduce considerable computation and communication overheads. In contrast, physical-layer authentication reduces repeated cryptographic processing, yet many existing designs rely on a single authentication tag or fragment, making legitimate messages vulnerable to false rejection when the authentication evidence is corrupted by fading, noise, Doppler effects, or imperfect signal recovery. To address this limitation, this paper develops a threshold-resilient cross-layer authentication method for secure vehicular communication. The method begins by forming a trusted communication session between the vehicle and the roadside unit through public-key-infrastructure-based mutual authentication. For each subsequent safety message, a message-and-session-bound authentication secret is generated using the session key, session nonce, vehicle pseudonym, message payload, and timestamp. This secret is then divided into multiple threshold shares, where each share is protected by a dedicated share-wise tag and embedded at the physical layer. At the receiver side, authentication succeeds only when at least t out of n share-tag pairs are correctly recovered and verified, followed by successful reconstruction of the message-bound secret. Formal and informal security analyses demonstrate resistance to replay, impersonation, message alteration, share-forgery, man-in-the-middle, and partial authentication-fragment corruption. The obtained performance results indicate that the proposed method preserves low computational overhead and moderate communication cost while enhancing authentication robustness when part of the physical-layer evidence is lost or corrupted.
Cardiovascular diseases constitute a significant and enduring concern within preventive cardiology, where Timely identification is critical for improving patient outcomes. Tradi-tional diagnostic methods often rely on costly instrumentation and specialist knowledge, restricting their utility in low-resource environments. Informed by contemporary de-velopments in medical deep learning, this study proposes Audio-CardioNet, an innova-tive hybrid model for identifying CVDs from phonocardiogram data. The architecture synthesizes three principal components: a multi-scale convolutional neural network for hierarchical temporal feature extraction; a bidirectional LSTM augmented with a Dual-Path Attention Mechanism to capture sequential and inter-feature relationships; and an Ant Colony Optimization algorithm for feature selection, achieving over 60% dimensionality reduction. Evaluation on a multi-source dataset of 15,420 PCG recordings demonstrated superior performance, with an accuracy of 98.2%, sensitivity of 98.5%, specificity of 97.8%, and an F1-score of 0.981, exceeding current benchmark methods. The framework also exhibits sub-50ms latency, indicating suitability for real-time, point-of-care applications however, this preliminary research-stage framework requires prospective clinical vali-dation before any consideration of deployment. Detailed per-class analysis, confusion matrices, and ablation studies substantiate the findings and facilitate clinical translation.
