Neuroimaging plays a key role in understanding the structural and functional brain changes associated with Alzheimer's disease (AD), especially in multimodal pattern recognition. However, single-modality approaches often fail to capture the complex nature of disease progression, while many multimodal methods suffer from high computational cost, overfitting, and limited interpretability. To address these challenges, we propose an explainable multimodal learning framework that integrates structural Magnetic Resonance Imaging (MRI) and functional Positron Emission Tomography (PET) data for accurate and transparent AD diagnosis. The proposed pipeline introduces Multimodal Optimized Diffeomorphic Anatomical Registration (MoDAR) for consistent cross-modality alignment. Feature extraction and fusion are performed using a Soft Progressive Attention–Enclosed Sparse Probabilistic Variational Autoencoder (SPA-SPVA), which learns compact and discriminative representations. For disease staging, we developed a Convolutional Masked Transformer (CoMFormer) that captures both local imaging biomarkers and global dependencies. Model parameters are optimized using a sine-chaotic Greylag Goose Optimization Algorithm (SC-GGOA) strategy to ensure stable convergence. Shapley Additive exPlanations (SHAP) are employed to provide interpretable insights into model decisions. Experimental results on the AD Neuroimaging Initiative (ADNI) dataset demonstrate an accuracy of 98.14%, outperforming recent methods. The framework is generalizable and can be extended to other multimodal biomedical data, highlighting its potential for real-world healthcare applications.

Early and accurate diagnosis of Alzheimer's Disease (AD) is critical for effective disease management and progression delay. Researches have been done in past towards better study of Alzheimer's, but advancements in feature engineering-cum-learning methodologies have still created scope to overcome the limits of previous methods and achieve more accurate modelling and classification. Here, we propose a novel model, LEFF-ViT (Locally Enhanced Feedforward Vision Transformer), for AD classification along with a framework culminating an idea of using separate segmented brain subregions as a marked feature engineering element. For this Segmentation of MRI images are done to extract White Matter (WM), Gray Matter (GM), and Cerebrospinal Fluid (CSF) regions using a Deep Residual Squeeze-Inception U-Net (De-RIS U-Net). Subsequently, a novel DWFE-Net is employed to extract discriminative spatial features. Finally, LEFF-ViT integrates a Vision Transformer with Multi-Head Self-Attention and a Locally Enhanced Feedforward Network (LFFN) to effectively capture both local and global contextual information for accurate classification. The experimental results demonstrate that the proposed model achieves an accuracy of 98.68 %, a sensitivity of 96.45 %, a specificity of 98.17 %, a Dice score of 96.36 %, and a Jaccard index of 92.31 %, which nearly outperforms the existing state-of-the-art methods across multiple evaluation metrics.

Sleep is beneficial to the restoration of the mind and general brain health. Multi-channel electroencephalography (EEG) systems are the traditional method of sleep analysis, but are costly, technologically demanding, and impractical for real-world use. This paper presents an end-to-end classification framework using self-recorded single-channel wearable EEG data collected under real-world resting conditions, capturing natural inter-session variability and noise characteristics of practical wearable acquisition. Eight spectral power features were extracted across Delta, Theta, Alpha1, Alpha2, Beta1, Beta2, Gamma1, and Gamma2 bands. Four classifiers Support Vector Machine, Random Forest, Multi-Layer Perceptron and XGBoost were evaluated under both multivariate and univariate band-wise settings. To ensure reproducibility, 5-fold stratified cross-validation was applied across all models using leakage-free pipelines. XGBoost and Random Forest achieved the highest CV accuracy of 82.49Dataset will be made available on reques% (± 1.75%) and 81.85% (± 1.59%) respectively, with ROC-AUC of 0.891 for both ensemble models. Univariate analysis revealed that gamma-band activity alone achieves 81.5% accuracy, identifying it as the dominant spectral indicator of post-sleep cognitive recovery. These results demonstrate the feasibility of low-cost single-channel EEG for practical sleep-state classification.

The growing adoption of EEG-enabled consumer Internet of Medical Things (IoMT) devices and wearable neuro-monitoring systems in healthcare and home-centric environments has raised critical concerns regarding data privacy, trust, and regulatory compliance. Although centralised deep learning approaches achieve strong performance in EEG-based emotion and cognitive state recognition, their reliance on centralised data aggregation limits scalability and applicability in privacy-sensitive consumer healthcare settings. Furthermore, existing federated learning approaches often inadequately address trust heterogeneity, non-IID neurophysiological variability, unreliable client participation, and the emerging requirement of federated unlearning in distributed IoMT environments. This paper proposes FedTrust-UNLEARN, a trust-aware federated learning framework with auditable unlearning for privacy-preserving EEG intelligence across heterogeneous consumer IoMT networks. The proposed framework integrates adaptive spectral representation learning, dynamic trust-aware aggregation, and an auditable federated unlearning mechanism that enables selective attenuation of client influence without requiring complete retraining. Extensive experiments on the SEED, SEED-IV, and DREAMER datasets under subject-dependent and subject-independent settings demonstrate competitive performance relative to centralised learning while improving robustness against unreliable clients, privacy preservation, and practical unlearning behaviour. The results demonstrate stable collaborative optimisation, bounded post-unlearning degradation, and communication-efficient deployment, highlighting the practicality of trustworthy, regulation-aware federated EEG intelligence for real-world consumer IoMT healthcare applications.

Alzheimer's Disease (AD) has emerged as one of the serious causes of death worldwide effecting the elderly population. As the development of this disease is related to neuronal loss in brain sub regions, the focus of this paper is on segmenting the brain into sub regions like GM (Grey Matter), WM (White Matter) and CSF (Cerebrospinal Fluid) followed by AD classification. This paper presents a complete framework for AD classification using feature maps of segmented tissue of brain. Two novel approaches have been proposed for classification task, where in first approach Self-Augmenting CNN and in the second approach Modified ResNet model has been used. Segmentation model achieves the highest similarity score of 99.53% for DS (Dice Similarity) and 99.09% for IOU (Intersection of Union) on GM segment. For Multi-class classification, accuracy of 97.92% is achieved with Self-Augmenting CNN, while Modified-ResNet model achieves 98.75% accuracy on segmented feature maps.

Alzheimer’s disease (AD) is one of the most severe kinds of dementia that affects the elderly population. Since this disease is incurable and the changes in brain sub-regions start decades before the symptoms are observed, early detection becomes more challenging. Discriminating similar brain patterns for AD classification is difficult as minute changes in biomarkers are detected in different neuroimaging modality, also in different image projections. Deep learning models have provided excellent performance in analyzing various neuroimaging and clinical data. In this survey, we performed a comparative analysis of 134 papers published between 2017 and 2022 to get 360° knowledge of the AD kind of problem and everything done to examine and deeply analyze factors causing this. Different pre-processing tools and techniques, various datasets, and brain sub-regions affected mainly by AD have been reviewed. Further deep analysis of various biomarkers, feature extraction techniques, Deep learning and Machine learning architectures has been done for the survey. Summarization of the latest research articles with valuable findings has been represented in multiple tables. A novel approach has been used representing classification of biomarkers, pre-processing techniques and AD detection methods in form of figures and classification of AD on the basis of stages showing difference in accuracies between binary and multi-class in form of table. We finally concluded our paper by addressing some challenges faced during classification and provided recommendations that can be considered for future research in diagnosing various stages of AD.

Alzheimer's disease is a type of progressive neurological disorder which is irreversible and the patient suffers from severe memory loss. This disease is the seventh largest cause of death across the globe. As yet there is no cure for this disease, the only way to control it is its early diagnosis. Deep Learning techniques are mostly preferred in classification tasks because of their high accuracy over a large dataset. The main focus of this paper is on fine-tuning and evaluating the Deep Convolutional Networks for Alzheimer's disease classification. An empirical analysis of various deep learning-based neural network models has been done. The architectures evaluation includes InceptionV3, ResNet with 50 layers and 101 layers and DenseNet with 169 layers. The dataset has been taken from Kaggle which is publicly available and comprises of four classes which represents the various stages of Alzheimer's disease. In our experiment, the accuracy of DenseNet consistently improved with the increase in the number of epochs resulting in a 99.94% testing accuracy score better than the rest of the architectures. Although the results obtained are satisfactory, but for future research, we can apply transfer learning on other deep models like Inception V4, AlexNet etc., to increase accuracy and decrease computational time. Also, in future we can work on other datasets like ADNI or OASIS and use Positron emitted tomography, diffusion tensor imaging neuroimages and their combinations for better result.

Alzheimer's disease (AD) is a form of dementia that is irreversible in nature with no effective cure till date. It ranks seventh in terms of causing mortality, predominantly impacting the elderly demographic. Therefore, its early diagnosis is an important concern for controlling the progression of this disease. This paper aims to design a framework for classifying various stages of this disease using brain MRI. Nowadays, deep learning approaches has gained much attention due to its promising results in object detection and classification tasks. But the requirement of huge dataset is the most common issue with these architectures. To overcome this issue concept of Transfer Learning (TL) is adopted by many researchers to take the advantage of pre-trained models. In this work TL has been applied by fine tuning the EfficientNetB0 model on Kaggle dataset that classifies the four stages of Alzheimer's disease. The obtained result proves that the model outperforms the state-of -the-techniques achieving an accuracy of 95.78% for multi-class classification.

Activity recognition is a challenging and practical research problem in computer vision. A good number of applications are approached by researchers in this area, like activity monitoring in homes, smart healthcare systems, surveillance & monitoring for security, and more. Literature around this problem provides some solutions where machine learning approaches have profoundly been utilized. This paper proposes a method based on deep neural networks to classify and recognize human activities for video surveillance applications. In order to achieve our objective, we ensemble the two different specialized neural network models, viz convolutional neural network (CNN) and recurrent neural network (RNN), to perform activity classification. Here, CNN extracts relevant spatial features from the video sequence and RNN creates a temporal relationship among those spatial features, which in turn leads to activity classification. To evaluate the effectiveness of the proposed method, we use precision, recall and accuracy metrics. On evaluation, the method is found to be more efficient in classifying video sequences, with an accuracy of up to 91.90%.