Cardiovascular diseases remain a leading cause of mortality worldwide, necessitating accurate and early detection strategies. Emerging evidence suggests a strong correlation between retinal microvascular alterations and systemic cardiovascular conditions, including heart disease and hypertension. This study leverages retinal fundus imaging as a non-invasive biomarker for cardiac health assessment. Utilizing the kaggle consolidated retinal image dataset, we propose a deep learning-based approach employing an ensemble of ResNet50V2 and EfficientNet4 to analyze morphological changes in retinal blood vessels indicative of cardiac risk. The proposed model achieves an accuracy of approximately 97.5% in predicting the likelihood of heart disease risk, demonstrating the potential of ocular biomarkers in cardiovascular risk stratification. This research underscores the utility of integrating medical imaging and deep learning to develop accessible, cost-effective screening tools for early heart disease risk prediction.
In agriculture, detection of plant diseases is a crucial activity that is important for ensuring crop health and increasing productivity. In this project, we provide an in-depth analysis of the use of a Convolutional Neural Network (CNN) for plant disease detection. This paper utilizes a dataset that includes a sizable number of photos showing both healthy plants and plants with various diseases. The dataset is enhanced and pre-processed to improve the ability to be generalized the model. Convolutional layers, pooling layers, and fully connected layers are used in the creation and training of a CNN architecture to extract useful characteristics from the images. The model's parameters are optimized during the training phase using stochastic gradient descent and a properly selected learning rate. Over-fitting is avoided by using regularization techniques like dropout and weight decay. The proposed model demonstrates strong performance, achieving an accuracy of 96.84% in classifying plant diseases from leaf images. The dataset used for training and evaluation was obtained from Kaggle and includes images representing multiple plant disease categories. By leveraging deep learning techniques, the model effectively distinguishes between healthy and diseased leaves, addressing limitations associated with traditional visual inspection methods, which are often time-consuming and prone to human error. The findings highlight the potential of automated disease detection systems in agriculture, enabling early diagnosis and timely intervention to reduce crop losses and enhance overall crop management practices.
The present study aimed to develop an efficient one-pot synthetic strategy for Chromen–imidazole hybrid derivatives and to evaluate their potential as antibiofilm and antifungal agents using integrated experimental and computational approaches. A series of nine chromen–imidazole hybrids (4a–i) were synthesized via a one-pot multicomponent reaction catalyzed by oxalic acid, employing benzil, substituted aldehydes, substituted amines, and ammonium acetate as precursors. The synthesized compounds were structurally characterized using IR, NMR, and mass spectrometry. All derivatives were systematically evaluated for their antimicrobial, antifungal, and antibiofilm activities, while their pharmacokinetic profiles were predicted using ADMETLAB 2.0. In addition, molecular docking and molecular dynamics simulations were performed to assess binding affinities and the stability of ligand–protein complexes, focusing on key molecular targets identified through network pharmacology. The developed synthetic protocol proved to be simple, efficient, and versatile, affording structurally diverse hybrids with significant biological activities. Compounds 4c, 4h, and 4i exhibited superior biofilm inhibitory activity against Pseudomonas aeruginosa compared to ciprofloxacin, whereas compounds 4d, 4h, and 4i showed notable antifungal activity against Candida albicans. Network pharmacology analysis identified SRC and HSP90AA1 as key targets associated with microbial biofilm formation and stress response pathways. Molecular docking revealed strong binding affinities, particularly for compounds 4g and 4e, while molecular dynamics simulations confirmed stable ligand–protein interactions. Furthermore, ADME predictions indicated favorable pharmacokinetic properties. Overall, these findings highlight the potential of chromen–imidazole hybrids as promising multifunctional therapeutic candidates.
Single crystals of ferroelectric guanidinium uranyl sulfate tri-hydrate (GUS3H) were grown with vanadyl ion doping by slow evaporation technique at room temperature. EPR spectroscopic study revealed two distinct sites for the impurity vanadyl ion in the host crystal. It was confirmed that the vanadyl entered at the two uranyl sites and local relaxations in the coordination spheres were caused around vanadyl ions. The two vanadyl impurity complexes are identical in chemical coordination but they are different crystallographically. Spin Hamiltonian analysis revealed axial symmetry of the guest vanadyl complexes with g|| =1.929±0.001; g┴ = 1.987±0.001 and A|| = 202± 5 G, A┴ = 82± 5G .The bonding parameters suggested that the in-plane covalency due to σ-bonding is weaker than the out-of-plane covalency due to π-bonding for the guest vanadyl complexes .The EPR features are almost temperature independent between RT and LNT and there is no indication of any phase transition between these temperatures.
Thermal Barrier Coatings (TBCs) are pivotal in safeguarding high-temperature components in aerospace and energy applications. This paper explores the structural composition of TBC systems, including substrate, bond coat, thermally grown oxide (TGO) layer, and ceramic topcoat. Various deposition techniques such as Atmospheric Plasma Spraying (APS) and Electron Beam Physical Vapor Deposition (EB-PVD) are discussed for their impact on coating performance. Challenges such as crack propagation, thermal fatigue, and environmental degradation are analysed to understand their influence on TBC longevity. Notably, the potential of advanced materials, including rare earth zirconates and hafnium-based ceramics, is evaluated to enhance thermal stability and damage tolerance. The ongoing research aims to mitigate existing limitations while pushing the boundaries of TBC technology in high-temperature applications, ensuring improved performance and efficiency
Long-sequence multivariate time series forecasting is widely applicable across various real-world scenarios. The Vanilla transformer has recently been used to predict the long-sequence time series. Transformers are initially designed for tasks like machine translation, where the order of the input sequence is crucial, but the specific temporal dependencies of the time series may not be fully captured. Moreover, its autoregressive nature poses limitations. This paper proposes a novel approach named FANT (frequency-attention-based non-autoregressive transformer) tailored for accurately forecasting long-sequence multivariate time series. The FANT method decomposes the time series into trend and seasonality components and enriches the input data with a calculated time vector using TimetoVec. This augmentation aids in capturing both short-term and long-term dependencies within the time series, facilitating the model’s comprehension of intricate temporal relationships. The approach employs the Fourier transform to shift the data into the frequency domain, where self-attention is subsequently calculated to discern patterns and periodicity within the data. Importantly, the FANT model is non-autoregressive, enabling simultaneous generation of all sequence elements. This characteristic minimises error propagation and reduces computational time, enhancing overall efficiency. The empirical evaluation of the FANT model encompasses seven datasets from financial and energy domains. Results indicate its superiority over the vanilla Transformer, showcasing a minimum relative improvement of 18.74% across all datasets. Additionally, an ablation study underscores the effectiveness of the frequency-attention model compared to self-attention.
Automatic music transcription (AMT) is still a major challenge in music information retrieval. Polyphonic audio is a mix of several overlapping frequencies, making it challenging for the usual monophonic pitch estimators. Therefore, models like CREPE perform well on single voice audio. But when they encounter polyphonic mixes or low SNR signals, they perform poorly. This paper addresses the stated problem with a modular hybrid framework to make AMT more robust. It uses neural source separation along with a dual-branch feature-guided gating mechanism. The pipeline begins with demixing the polyphonic audio into different constituent stems vocal, bass and other instruments. This is achieved using Demucs a state-of-the-art source separation model. These stems simultaneously pass through two branches one after another, with one branch being a pitch-tracking branch that runs torchcrepe to get candidate F0 contours, while the other pulls out a voicing mask from high-resolution Mel-spectrograms and chroma-energy features. The fusion of pitch estimates from torchcrepe equipped branch with voicing mask from the other branch helps eliminate false activations like background noise. Also, since all the stems go through the above steps, ultimately, we obtain the MIDI representations of all the instruments and voices involved in the audio. This approach is tested on Annotated-VocalSet and Vocadito datasets and found that there is significant reduction on false alarms caused by noise and a boost in raw pitch accuracy resulting in refined multi-track MIDI representation.
Knee osteoarthritis (KOA) is a widespread degenerative joint disease that affects millions globally. While manual interpretation of X-ray images remains common in clinical practice, it is time-consuming and prone to subjective errors. In this study, we propose a hybrid automated diagnostic model for KOA detection using feature extraction from Convolutional Neural Networks (CNNs) combined with traditional feature selection methods. The hybrid Feature Selection (FS) strategy includes Mutual Information (MI), Linear Discriminant Analysis (LDA),Analysis of Variance(ANOVA), and K-means clustering. In addition, advanced deep learning architectures such as ResNet and EfficientNet were incorporated for benchmarking purposes to evaluate the effectiveness of the proposed hybrid model. To enhance the robustness of the proposed model , advance deep learning architectures were also considered for comparative analysis. The model received training and validated using a publicly available X-ray dataset containing 3,815 images obtained from the Kaggle repository. After preprocessing and balancing the dataset, we extracted deep features via a retrained CNN and selected the most informative features using the hybrid (FS) strategy. The proposed model achieved a prediction accuracy of (98%), The novelty of this work lies in integrating deep learning with multiple statistical feature selection techniques within a unified hybrid framework for early KOA detection accuracy and reducing diagnostic workload but also enhances model generalizability and robustness. The study also acknowledges limitations such as dataset imbalance and lack of external validation.
Noncoding peptides (ncPEPs), which are small peptides derived from noncoding RNAs, have gained immense attention in recent years because of their diverse biological functions and their specific expression. Recent advancements in technologies such as ribosome profiling and mass spectrometry have provided valuable insights into the coding potential of noncoding RNAs and their involvement in encoding micropeptides. These peptides are involved in crucial biological roles, including gene expression, cellular communication, metabolism and immune system regulation. Furthermore, they are also linked to various disease conditions, including cancer, making the study and characterization of ncPEPs essential for understanding their biological importance. This study focuses on analyzing a large dataset of ncPEPs from the SPENCER database to explore structural features and amino acid composition, with a specific focus on posttranslational modification sites. ncPEPs derived from 15 different cancer types were retrieved, with a subset of 4,505 unique sequences retained for detailed analysis. All sequences, ranging from 7 to 52 amino acids, were used for amino acid composition analysis, with a subset of peptides between 15 and 52 amino acids further subjected to structure prediction using AlphaFold2 and AlphaFold3. A particular focus was placed on serine, threonine and tyrosine residues as key sites of posttranslational modification.
Signatures remain a primary authentication method in banking, especially for cheque verification. However, automated signature detection faces challenges due to diverse cheque layouts, overlapping text fields, and handwriting variability. This paper proposes an explainable signature detection framework, Optimized IoU-Aware YOLOv8 with Bidirectional Feature Pyramid Network (OYO-BiFPN), designed for robust and interpretable cheque analysis. The framework employs Edge-Preserving Weighted CLAHE and morphological refinement for stroke enhancement, coupled with Confidence-Guided Dual-Branch Triplet ResNet for reliable feature extraction. An OCR-guided filtering stage eliminates false positives from printed text, while Stellar Oscillation Optimizer (SOO) ensures optimal parameter tuning. To enhance interpretability, Grad-CAM++ provides visual insights into detection decisions. Experimental results on benchmark cheque datasets demonstrate superior performance with 99.8% accuracy, 99.2% precision, and 99.8% recall, confirming the model’s reliability and transparency. The proposed OYO-BiFPN framework advances secure banking automation by integrating accuracy, robustness, and explainable AI principles in signature verification.
Autism spectrum disorder (ASD) is a lifelong neurodevelopmental disorder. ASD patients have communication problems. The screening of ASD is both time-consuming and requires a specialist to conduct an analysis. This involves restriction in the diagnoses, which are delayed, and the age at which it is normally detected is 5 years. In recent years, accurate and early screening has been performed using machine learning (ML) and deep learning (DL). These approaches have potential to study patterns that are related to ASD as far as use of neuroimaging, physiological, and Behavioural data sets is concerned. This literature review focuses on the ML and DL approaches applied in different works, such as the support vector machines (SVM), random forests (RF), logistic regression (LR), Decision Tree (DT), convolutional neural networks (CNN), and graph neural networks (GNN). It also examines the new role of hybrid and multimodal integration strategies. Various studies have shown a high accuracy of more than 90%, especially with deep neural networks. Small and unbalanced data, gender bias, and insufficient external validation are some of the challenges identified in this review. Altogether, ML and DL techniques provide higher accuracy and speed of ASD diagnosis. Standardized datasets, cross-site validation, and interpretable models should be prioritized in future studies in order to come up with effective and generalizable diagnostic tools.
This paper presents a novel application of BERT-driven contextual topic modeling to enhance the analysis of financial news, focusing on Indian financial data from 2003 to 2025. Traditional topic modeling approaches, such as Latent Dirichlet Allocation (LDA), treat words as isolated units, limiting their ability to capture semantic context. Leveraging BERT’s bidirectional transformer architecture, this study generates context-aware embeddings that improve the coherence and interpretability of extracted topics. The proposed framework employs BERTopic, integrating sentence-transformers for semantic embeddings, Uniform Manifold Approximation and Projection (UMAP) for dimensionality reduction, Hierarchical Density-Based Spatial Clustering of Applications with Noise (HDBSCAN) for clustering, and class-based Term Frequency-Inverse Document Frequency (c-TF-IDF) for topic representation. This methodology enables the identification of latent financial themes, including banking, currency fluctuations, regulatory norms, credit growth, and institutional governance, providing actionable insights for investors and policymakers. Visualization techniques, such as intertopic distance maps, word clouds, and heat maps, facilitate exploration of semantic relationships and thematic clusters within the news corpus. Experimental results demonstrate that BERT-based topic modeling outperforms traditional methods by uncovering semantically coherent and non-overlapping topics, thereby enabling a more transparent and fine-grained understanding of complex financial narratives. This approach supports timely detection of market trends, risk factors, and sentiment shifts, contributing to improved decision-making within financial ecosystems.
The proper delineation of the multi-modal MRI of the gliomas and of the brain is important in diagnosis, surgical planning, and monitoring of treatment in neuro-oncology. There is no doubt that convolutional neural networks CNNs) have become the standard of medical image segmentation. But still there is a limited amount of comparative evidence on how different architectures model tumor morphology. This paper is a systematic evaluation of eight CNN-based segmentation models namely U-Net, ResU-Net, Attention U-Net, RAU-Net, Feature Pyramid Network (FPN), SegNet, BiSeNet and DeepLabV3+ using the BraTS 2020 dataset. Each model was trained with the same preprocessing and hyperparameter settings, without augmentation or pre-training, to determine architecture-based effects. Performance was evaluated on a variety of accuracy measures (mean Intersection over Union, precision, recall, specificity, AUC) and efficiency measures (inference time). Consistently, FPN and ResU-Net outperformed ground truth (mIoU ≈0.896), strong recall (>0.92), and sub-milliseconds inference time, indicating better ability to detect infiltrative tumor margins and still be clinically deployable. Attention U-Net presented the best accuracy (0.96), minimizing false positives, but lightweight BiSeNet did not capture tumor boundaries despite being able to run fast. This work contributes insights into the encoding of glioma morphology by deep learning models, and identifies directions to pursue biologically-informed model development, by connecting the architectural performance of the deep learning model to the clinically-relevant error profiles.
Understanding and classifying musical instrument sounds is a foundational task in Music Information Retrieval (MIR) and supports applications such as automated transcription, digital music organization, and audio content analysis. Distinguishing between instruments requires models capable of recognising complex timbral characteristics from audio signals. The paper under discussion explores the problem of classifying piano and violin notes by supervision with machine-learned algorithms over the features of the NSynth dataset and the additional dataset, where users record their rhythms (RACK). The features of audio (harmonic and percussive elements, Mel-Frequency Cepstral Coefficients (MFCCs), Mel spectrogram features, chroma statistics and spectral contrast) were obtained with the help of the LibROSA library. Four trained and optimised supervised classifiers were trained on Scikit-learn (Random Forests, AdaBoost, Support Vector Machines, and k-Nearest Neighbours). Findings show that Support Vector Machines performed best at 99.84% on the NSynth dataset, where the other two algorithms (Random Forests and AdaBoost) outperformed the performance metrics. Test on the self-captured RACK dataset was found to perform worse because of the variability of note characteristics, but accuracy was improved significantly when irregular short violin samples were eliminated. Altogether, the research illustrates the efficiency of spectral contrast and MFCC features in instrument recognition, as well as shows the high potential of supervised learning methods that can be used to obtain reliable audio classification.
The sharded PBFT protocol is proposed in this study as a solution to the problems of low transaction throughput, low scalability, and excessive energy consumption. Sharded-PBFT allows node scalability to be linear by partitioning the network into clusters. Throughput is increased by the transactions in each cluster being executed in parallel. Sharded-PBFT's leader selection procedure makes it more energy-efficient than conventional PBFT. Metrics including data transmission, processing time, consensus latency, block congestion, and message exchange are used to verify Sharded-PBFT's performance. It demonstrates that PBFT transmits 4.6 times more data than Sharded-PBFT with different nodes. For different nodes and block sizes, there is a 48% decrease in the number of messages sent and a 2.17 and 1.7-times reduction in processing time, respectively. The decentralized, tamper-resistant nature of blockchain technology is founded on a peer-to-peer network has led to its widespread usage in financial applications and its recent expansion to industrial applications. The well-known conundrum of blockchain scalability, security, and decentralization, however, significantly hinders the widespread use of traditional Blockchain technology because of its poor scalability. With respect to this matter, sharding—the division of a network into several shards in order to minimize the cost associated with repeating computation, storage, and communication in each complete node— is among the most viable methods to attain horizontal scalability in conjunction with the growing size of the network.
Lung squamous cell carcinoma (LUSC) lacks recurrent mutation-driven therapeutic targets and is characterized by profound metabolic reprogramming. Although fibroblast growth factor receptor 3 (FGFR3) has been implicated in several malignancies, its functional relevance in LUSC remains incompletely defined. In this study, we performed an integrative multi-omics analysis using TCGA transcriptomic and mutational datasets to determine whether FGFR3 dysregulation in LUSC is mutation-driven or expression-driven, and to evaluate its association with glycolytic activation. Differential expression analysis demonstrated significant upregulation of FGFR3 and SLC2A1 (GLUT1) in LUSC tumors compared with adjacent normal tissues. In contrast, somatic FGFR3 alterations were infrequent (~3%) and lacked recurrent hotspot enrichment. Correlation analysis revealed a significant positive association between FGFR3 and multiple glycolysis-related genes, particularly SLC2A1. Gene set variation analysis confirmed increased glycolysis pathway activity in FGFR3-high tumors. Functional enrichment and protein–protein interaction analyses demonstrated convergence of FGFR3-associated genes on central carbon metabolic pathways. Protein-level validation using Human Protein Atlas data confirmed heterogeneous FGFR3 expression and robust SLC2A1 overexpression in tumor tissues. Collectively, these findings indicate that FGFR3 dysregulation in LUSC is predominantly expression-driven and transcriptionally coordinated with SLC2A1-mediated glycolytic activation, defining a signaling–metabolic axis that may support metabolic stratification and future therapeutic exploration.
Many factors, such as increasing civilian use and military employment of small unmanned aerial vehicles (UAVs), have resulted in an increase in demand for reliable detection methodologies that can be used at low altitudes and under cluttered conditions. Micro-doppler-based sensing using radar-based systems provides useful motion-related information, but there are many challenges in utilizing them due to background noise and weak signal levels. This paper explores an optimised neural network framework with reduced complexity inspired by MobileNetV2 for the detection and categorizaon of small unmanned aerial vehicels based on micro movements produced by the propeller.Spectrograms produced from the DIAT-Usat dataset. The original dataset was enhanced through augmentations (rotation, horizontal flip, vertical flip, and brightness variations). Raw radar signals were transformed into time-frequency representations, which were analyzed using several different CNN architectures to determine classification performances. Comparative experiments, including EfficientNet-b0, Densenet-121, Xception, Resnet-50, and VGG 16, demonstrated that the proposed model exhibited rapid convergence and stable generalization towards compactly represented spectrograms. With a training accuracy of 98.9%, a validation accuracy of 98.2%, and a minimal validation loss, it appears that a well-designed optimised convolutional network may effectively identify target specific movemt patterns and thus is applicable for real-time aerial surveillance of small unmanned aerial vehicle operations within diverse environmental conditions.
Breast cancer is one of the leading causes of mortality among women worldwide, making early and accurate diagnosis essential for effective treatment and improved survival rates. Mammography is widely considered the standard screening technique for detecting early signs of breast cancer; however, manual interpretation of mammograms may lead to diagnostic errors, resulting in missed detections or unnecessary treatments. To address these challenges, this study proposes an enhanced machine learning–based framework for accurate breast cancer diagnosis using mammographic images. The proposed approach utilizes deep learning techniques, particularly Convolutional Neural Networks (CNNs), to automatically learn discriminative features from mammograms. Several state-of-the-art architectures, including VGG19, ResNet50, InceptionV3, DenseNet121, and MobileNetV2, are evaluated to analyze their effectiveness in breast cancer classification. The framework incorporates multiple optimization strategies such as transfer learning, class weighting to handle imbalanced datasets, and variations in input image sizes to improve model performance. Experimental evaluation is conducted using the CBIS-DDSM and mini-MIAS mammogram datasets. Among the tested models, the MobileNetV2 architecture combined with transfer learning from ImageNet and additional pre-trained weights demonstrates superior performance, achieving an accuracy of 87.08% on the CBIS-DDSM dataset. Furthermore, separating mammogram samples into masses and calcifications and applying class weights contribute to improved classification performance. The results are compared with existing approaches, demonstrating the effectiveness of the proposed enhanced machine learning framework for breast cancer detection in mammographic images.
Multiple technologies are being developed daily to support farmers and facilitate smart agriculture,. One critical practice for detecting whether a disease infects a plant is to observe patterns, often visible on leaves as spots of different colors. Recently, machine learning and image processing techniques have emerged to detect plant diseases based on the symptoms appearing on leaves and stems. Extensive research is carried out in agricultural image processing. In some cases, research outcomes have achieved results surpassing human vision capabilities. By implementing these technologies as user-friendly applications, farmers can easily detect crop diseases at an early stage. This will help avoid disease spread and maximize yield. The occurrence of plant diseases is mainly due to nutrient deficiencies and various environmental conditions. Such deficiencies impede normal growth, resulting in production loss and poor yield quality, since primary nutrients play a vital role in optimal plant development. Addressing this, the present study focuses on nutrient analysis in cotton crops using a novel Convolutional Neural Network framework as a multi-output regression model. A dataset of 1,000 real-time images collected from cotton fields for model training and evaluation. Once nutrient deficiencies are estimated, the model suggests possible organic fertilizers to the farmer. The model achieves a test loss of 1.16, a test MAE of 1.51, and a test MSE of 9.76.
Varicocele remains the most prevalent correctable finding in male infertility, yet its clinical interpretation is increasingly challenged by the limited explanatory power of conventional semen analysis. Emerging evidence positions varicocele as a biologically active lesion characterized by redox imbalance and downstream genomic instability, rather than a purely anatomical venous disorder. Oxidative stress (OS) and sperm DNA fragmentation (SDF) have therefore gained translational relevance as mechanistic intermediates linking varicocele to impaired fertilization, defective embryogenesis, recurrent pregnancy loss, and suboptimal assisted reproductive outcomes. Mechanistically, varicocele induces a convergence of hyperthermia, venous stasis, hypoxia-like stress, inflammation, and mitochondrial dysfunction, culminating in excessive reactive oxygen species (ROS) generation and failure of redox homeostasis. Given the intrinsic vulnerability of spermatozoa, characterized by limited antioxidant defenses and minimal DNA repair capacity, this oxidative milieu promotes lipid peroxidation, chromatin destabilization, and DNA strand breaks. Clinical syntheses consistently demonstrate elevated SDF in men with varicocele and indicate that varicocelectomy can reduce oxidative damage markers and SDF in selected patients, supporting a biologically responsive axis rather than incidental association. Despite these advances, critical uncertainties persist regarding assay standardization, clinically actionable SDF thresholds, and the extent to which biomarker improvement translates into live birth benefit. Current guidelines recommend selectively using SDF in clinical decisions for unexplained infertility, recurrent pregnancy loss, and ART failure, shifting toward phenotype-driven care. The OS-SDF-varicocele axis moves male infertility from descriptive approaches to mechanism-based precision treatment. Progress relies on molecular phenotyping, validating outcome-linked biomarkers, and redefining ‘idiopathic’ infertility into biologically distinct groups.
