Maniruzzaman

Magnetic resonance imaging (MRI) is a critical clinical tool that requires precise and reliable interpretation for effective brain tumor diagnosis and timely treatment planning. Deep learning methods have advanced automated tumor classification greatly in the last few years, but many of the current methods are still challenged by a lack of interpretability, a lack of testing on region-focused data, and a lack of model robustness testing. Such limitations reduce clinical trust and limit the practice of automated diagnostic systems. To address these challenges, this study proposes an interpretable deep learning model for classifying brain tumors using the PMRAM dataset, which is a Bangladeshi brain MRI collection containing four categories: glioma, meningioma, pituitary tumor, and normal brain.. The proposed pipeline combines image preprocessing and feature enhancement methods, and then it trains a series of squeeze-and-excitation (SE)-enhanced convolutional neural networks such as VGG19, DenseNet201, MobileNetV3-Large, InceptionV3, and EfficientNetB3. The SE-enhanced EfficientNetB3 performed best, with 98.70% accuracy, 98.77% precision, 98.70% recall, and 98.70% F1-score. Cross-validation also demonstrated stable performance, with a mean accuracy of 96.89%. The model also exhibited efficient inference with low GPU memory consumption, enabling predictions in about 2–4 s per MRI image. Grad-CAM++ and saliency maps were used to improve the transparency of the results, and it was found that the network was concentrated on the clinically significant parts of the tumor, which affected the model predictions. Further robustness analysis and cross-dataset testing are additional evidence of the generalization possibility of the model. An online application was also implemented to allow real-time prediction and visual explanation of brain tumors. Overall, the proposed framework offers a precise, interpretable, and promising solution to automated brain tumor classification using MRI images.

Timely and proper identification of dengue is a key to successful clinical management, especially in resources-constrained environments. There is a problem of reliable and rapid cases of dengue in the background of symptoms similar to the other febrile diseases and dynamic hematological alterations. The paper introduces an optimized version of Random Forest analysis with LIME, which is used to predict dengue based on routine hematological variables. The model was trained and tested on data of 1,003 patient records, and SMOTETomek balancing was used to handle the issue of class imbalance. Its performance was excellent with a testing accuracy of 99.60 %, F1-score of 0.9960, Matthews Correlation Coefficient standing at 0.9908, and Cohen Kappa of 0.9980. Model robustness was realized by five-fold cross-validation with negligible variability. Analysis on computational efficiency revealed that the training and inference time were 1.037 s and 0.008 s, respectively, and thus it can be practically deployed in a clinical environment. ELIME-based explanations provided interpretable insights, highlighting low WBC count, platelet count, and PDW as key drivers of predictions. The proposed framework, unlike the current methods, offers a better level of accuracy and stability, as well as integrates interpretability and low computational cost, which is of paramount importance to the previous research. This paper shows that AI had a high-performing, explainable, and clinically relevant early detection of dengue.

The magnetic resonance imaging (MRI)-based classification of brain tumors is important in early diagnosis and treatment planning. In spite of that, current types of deep learning, such as convolutional neural networks (CNNs) and transformers, have some limitations, including imbalanced representation of classes, small differences between classes, high computational costs, and low interpretability, that make them clinically inapplicable. In order to address these constraints, we introduce a fast Tiny Vision Transformer, TinyViT is a lightweight hybrid model that combines convolutional feature extraction with transformer-based global context modeling to achieve fast and highly accurate brain tumor classification. The model has been trained and tested on the PMRAM dataset which contains four classes: glioma, meningioma, pituitary tumor, and normal. As the results of the experiments show, TinyViT is able to perform at the state of the art level featuring overall accuracy of 0.9971, F1-score of 0.9970, Matthews Correlation Coefficient (MCC) of 0.9961, and both PR-AUC and ROC-AUC values are 1.000. Moreover, inference times are minimal (train 5.46 s and 1.27 s per test batch) and show a good trade-off between computation and error rates. Grad-CAM images indicate regions impacted by tumors in agreement with expert radiological evaluation and improve interpretability and reliability. The results indicate that TinyViT is an accurate, computationally inexpensive, and clinically interpretable method to automatically classify brain tumors, and has high prospects of being clinically applied in medical image processing.

Amidst rising global food security challenges, early and precise detection of plant diseases has become essential—particularly for high-value crops such as mangoes. This study introduces a novel deep learning-based framework for the classification of mango leaf pathologies using advanced convolutional and hybrid neural architectures. A curated dataset of 800 high-resolution mango leaf images, collected from the Rajshahi region of Bangladesh, was preprocessed using extensive data augmentation and color space transformations to enhance generalization. Multiple models, including K-Nearest Neighbors, AlexNet, VGG16, VGG19, and EfficientNet-B7, were evaluated and compared against two proposed architectures: a custom Convolutional Neural Network (CNN) and a hybrid model integrating EfficientNet-B7, Long Short-Term Memory, and attention mechanisms. The proposed CNN model achieved 100% accuracy, precision, recall, and F1-score, outperforming all baseline models. The hybrid model achieved comparable results, demonstrating the effectiveness of combining spatial and temporal feature extraction for plant disease detection. Additionally, Grad-CAM visualizations provided interpretable diagnostic heatmaps, reinforcing the transparency and reliability of the model’s predictions. The proposed framework advances state-of-the-art agricultural diagnostics by offering a scalable, interpretable, and high-performing solution for real-time disease monitoring in mango cultivation. These findings hold strong potential for improving crop surveillance and addressing food scarcity in mango-producing regions.

The pervasive growth of Internet of Things (IoT) gadgets has exacerbated vulnerability of the network in that it necessitates real-time intrusion detection systems (IDS) that can function under rigid limits of computation. The standard methods of deep learning cannot be used in edge devices with limited resources, yet they are still accurate. In this paper, it is suggested that a lightweight ensemble machine learning model can be applied to the efficient intrusion detection of IoT devices by combining the Logistic Regression, the Random Forest, and the Gradient Boosting model with adaptive weight optimization. The feature selection based on mutual information is used to trim the set of features to 15 items and still achieve 98.3 percent of discriminative power to allow sub-10 milliseconds inferences. The framework assessed on NSL-KDD and TON IoT datasets can label the data with a 97.8% accuracy, a decrease in false positive by 26 percent and a generalization improvement of 1.4 percent. Training on regular CPUs only takes less than three minutes and thus can be deployed quickly and can be updated rapidly on edge gateways. These findings illustrate the possibility of scalable, energy efficient, and real time cybersecurity of IoT networks using the framework.

The growing power usage in clinical institutions, necessitated by unpredictable patient times and heavy usage of medical equipment, requires a high level of predictive models of sustainable operation. This paper is a machine learning forecasting approach to predicting the daily hospital energy demand incorporating environment parameters, occupancy data and equipment availability. The historical data were extensively preprocessed by outlier-detection, Min-Max-normalization and missing-value imputation. Recursive Feature Elimination was made to determine important predictors, which made it possible to compare the algorithms of Random Forest, Gradient Boosting, and Ridge Regression. Gradient Boosting had a better result of MAE of 0.103 kWh and R2 of 0.489 indicating that the gradient boosting is better suited to capture the temporal and operation dynamics. The analysis performed on a scenario basis showed that possible energy savings could be 12 percent by means of optimized scheduling of HVAC and load control. The model allows the preemptive resource distribution, which lowers the cost of operation and carbon intensity of healthcare infrastructure. Findings confirm the practicability of forecasting based on machines learning in real-time energy management, which underlies decarbonization agendas and sustainable operations of the facility in resource-restricted settings.

Chronic kidney disease (CKD) is a major health problem in the world today with the need to ensure that risks are stratified at an early age to avoid an irreversible loss of renal functions. The paper provides a machine learning architecture that combines extensive demographic, lifestyle, comorbidity, laboratory biomarker, medication, and environmental exposure information to predict CKD progression. Additional processing included median imputation, categorical encoding, outlier suppression and filtering use of variance. Dimensionality reduction used ANOVA F-test and principal component analysis in order to determine discriminative features such as serum creatinine, serum glomerular filtration rate, albumin-creatinine ratio, blood pressure indices, and electrolyte markers. Four algorithms, including the ridge regression, support vector machines, random forest, and gradient boosting were assessed by means of repeated stratified cross-validation. The gradient boosting had a better performance with an area under curve 0.867, accuracy 93.1, F1-score 96.3 and Brier score 0.048, which shows the best discrimination and calibration. The framework allows specific risk measurement that facilitates timely nephrology referral, individualized therapeutic intervention and evidence-based disease management planning in clinical practice.

The maintenance of chronic disease requires a continuous health checkup to prevent complications and minimize the number of times one goes to the hospital. The Internet of Things (IoT) is utilised to develop Remote Patient Monitoring (RPM) systems that assist doctors in gathering and reviewing health information in real-time. Although these systems have the potential to enhance care delivery, they are still affected by challenges such as delay of communication, poor security, system incompatibility and ineffective usage of the systems in the zones with weaker internet connections. The current paper introduces an IoT-based RPM system with six layers. It has Edge Artificial Intelligence (AI) to search promptly through anomalous health information, an impregnable blockchain to secure and conserve health records securely, and a digital twin design to assist in making forecasts about the future. This system was tested using a Python simulation mechanism in which sample health data were used. The performance indicated rapid identification of issues within 10.46 milliseconds and great data security based on blockchain. Figures and charts demonstrated that the system could sort out the normal and abnormal health data accurately and assist the physicians in making superior judgments. This design fixes most of the issues with the existing RPM system design and provides a secure, intelligent, and adaptable mode of chronic disease management.

Sleep disorders, including obstructive sleep apnea (OSA) and insomnia, have serious implications for cognitive health and quality of life. Conventional diagnostic techniques, such as polysomnography, are expensive and are not available to the majority of the population. Early detection of these disorders is essential for mitigating long-term health consequences, yet traditional diagnostic methods remain out of reach for many. In this paper, we present MB-CNN-LSTM-ATTN, a novel deep learning model combining Convolutional Neural Networks (CNN), Long Short Term Memory (LSTM), and attention mechanism to reliably diagnose sleep disorders. The model is employed to analyze the Sleep Health and Lifestyle Dataset, and the class imbalance problem is addressed using SMOTEENN, while feature selection is performed using Gradient Boosting. A classification accuracy of 97.65% was obtained by the model, and it exhibited better precision (96.87%), recall (98.13%), and AUC (98.93%) compared with the traditional classifiers. The experimental results demonstrate that the proposed model has a high performance, which makes it an acceptable and effective solution for automatic sleep disorder diagnosis and feasible in mobile health, wearable devices, access to be used clinically.