Central Tobacco Research Institute

This paper presents a robust hybrid artificial intelligence (AI) framework designed to improve diagnostic accuracy in medical imaging tests such as magnetic resonance imaging (MRI), computed tomography (CT), and positron emission tomography (PET) scans. The enhancement is achieved by combining deep learning methods with rule-based reasoning and model-agnostic explainability techniques. The proposed hybrid architecture integrates convolutional neural networks (CNNs) and transformer models for feature extraction, while incorporating expert-defined rule-based logic to strengthen interpretability and ensure consistency in decision-making. To improve transparency, model-agnostic explainability approaches such as SHAP (Shapley Additive Explanations) and LIME (Local Interpretable Model-Agnostic Explanations) are applied, offering deeper insights into AI-driven diagnoses. This integration addresses critical clinical concerns related to trust, traceability, and adoption of AI systems. Experimental evaluations conducted on benchmark medical imaging datasets achieved an accuracy of 94.2%, which is 3.8% higher than conventional deep learning approaches, demonstrating improved robustness. Additionally, the proposed system enhances clinical interpretability, with a 21% increase in explainability ratings provided by healthcare professionals. These findings highlight the novelty and clinical relevance of hybrid AI models by combining automation with decision support, thereby promoting greater trust and adoption of AI in diagnostic workflows across healthcare facilities.

Concrete compressive strength is a critical property that directly influences the durability and performance of concrete structures. This study develops a unified Machine Learning (ML) pipeline to accurately predict compressive strength using a benchmark dataset of material composition and curing parameters. The pipeline integrates robust preprocessing (missing value imputation, outlier removal, normalization), domain-specific feature engineering (e.g., water–cement ratio), and multiple ML models including Support Vector Machine (SVM), XGBoost, and CatBoost. Experimental results demonstrate that SVM achieves the highest classification accuracy (93.2%) and precision (93.4%), while ensemble models such as XGBoost and CatBoost yield superior generalization and ROC-AUC performance (0.9809 and 0.9803, respectively). This estimation ensures structural safety, optimizes material usage, reduces construction costs, and supports quality control, making it an essential step in sustainable and reliable infrastructure development.

The rapid growth of Internet of Things (IoT) devices has exposed weaknesses in network infrastructures, so efficient and effective cyberattack detection is absolutely important. The application of Quantum Machine Learning (QML), more especially Quantum Support Vector Machines (QSVM), to raise the detection accuracy of known cyberthreats in IoT environments is investigated in this work. To better manage the enormous dimensionality and complexity of IoT network traffic data than conventional approaches, QSVM uses quantum computing ideas including improved quantum feature encoding and kernel techniques. With an accuracy of 99.12% against 92.00%, along with improved precision, recall, and F1-score metrics, experimental data show that QSVM greatly beats the classical Support Vector Machine (SVM). An essential first step toward integrating quantum computing into practical cybersecurity applications, our results show the potential of quantum-enhanced classifiers to deliver strong, scalable, and highly accurate intrusion detection systems for IoT networks.

Effective hole deviation management is crucial for optimizing drilling operations, particularly in challenging geographic environments. This paper presents Deep Learning (DL) based DACRNET (Dynamically Adaptive Convolutional Regression Network) to address the formation of heterogeneity and drill string dynamics. The proposed approach integrates multiple stages: data acquisition, preprocessing, Exploratory Data Analysis (EDA), feature creation and transformation, data partitioning, and DACRNET model. The high-frequency drilling data, including measurements from sensors on drill strings and formation properties, are collected by the data acquisition process. Data preprocessing and EDA techniques visualize the trends and patterns, allowing for the identification of key variables influencing deviation. The feature creation and transformations enhance the dataset representation, facilitating improved learning performance. The dataset is then partitioned into training and testing subsets to ensure robust model evaluation. DACRNET, a DL model specifically designed for this task, leverages sequential and spatial patterns in the data to identify and predict deviations with high accuracy. This model is implemented using Python Software, from the evaluation, the minimized Mean Absolute Error (MAE) of $\mathbf{0. 0 2 7}$, Mean Square Error (MSE) of $\mathbf{0. 0 0 9 9}$, Root MSE(RMSE) value is $\mathbf{0. 0 9 9 4}$, and Mean Absolute Percentage Error (MAPE) of 4.69 are significantly attained. Predictions are used to guide real-time deviation management strategies, ensuring optimal drilling outcomes.

The online version contains supplementary material available at 10.1007/s12298-025-01570-w.

Feeding behaviour of pymetrozine-resistant (Pym-R) and pymetrozine-susceptible (Pym-S) brown planthopper Nilaparvata lugens Stal strains were investigated using standard honeydew and probing tests. Under both the treated and untreated conditions the food consumption of Pym-R strain was higher (4.0 and 12.4 cm2 area of honeydew respectively) compared to Pym-S strain (2.7 and 7.1 cm2, respectively). Probing activity of both the Pym-R and Pym-S strains increased on rice seedlings treated with pymetrozine (35 and 26 probing marks, respectively) compared to untreated seedlings (12 and 13 probing marks, respectively) and on treated seedlings the Pym-R insects probed more (35 probing marks) as compared to Pym-S insects (26 probing marks). Plausible reasons for differences in the feeding behaviour of pymetrozine resistant and susceptible populations is discussed.