Timely and accurate identification of brain tumors is crucial for optimizing patient outcomes, guiding surgical planning, and determining appropriate treatment strategies. Despite the advancements in MRI-based tumor detection and artificial intelligence, developing reliable and clinically applicable models remains challenging, particularly in contexts where robustness, interpretability, and consistency are critical. Existing approaches often lack advanced 3D imaging capabilities and robust Explainable AI (XAI) techniques, which limit their diagnostic utility and clinical adoption. To address these limitations, this study proposes a robust deep learning architecture that incorporates the improved EfficientNet B7, Xception, and ResNet152 architectures. We further proposed a novel classification model incorporating enhanced data augmentation and histogram equalization to automatically classify brain MRI images into four diagnostic categories. Three distinct datasets, one segmented dataset, and a merged dataset combining all sources were used to train and evaluate the models. The proposed models achieved accuracies in the range of 97–99%, demonstrating consistent and strong performance across datasets. To enhance diagnostic transparency, XAI techniques, including Grad-CAM and LIME, were employed to provide visual insights into the system’s decision-making processes. Additionally, this study incorporated a novel 3D reconstruction approach across the sagittal, coronal, and axial planes, providing a comprehensive view that supports improved diagnostic accuracy and precise clinical interpretation. Overall, this study demonstrates a novel unified framework integrating classification, segmentation, XAI, current gaps in brain tumor diagnostics and holding significant potential to enhance clinical reliability, transparency, and precision, ultimately improvinge clinical reliability, interpretability, and precision, ultimately supporting better patient outcomes.

• Proposed MVIC-Net which unifies four complementary ECG representations for robust sleep apnea detection. • Incorporated CINet architecture as a component of MVIC-Net that advances image-based ECG analysis via hierarchical interactive learning. • XAI ensures clinical transparency of CINet, aligning model decisions with physiologically meaningful signal regions. • Proposed MVIC-Net achieves 99.25% accuracy, setting a new benchmark for single-lead ECG-based apnea screening. • Multi-view fusion of MVIC-Net surpasses single-view and ensemble models; advantages confirmed statistically. • McNemar’s test highlights the significance of MVIC-Net’s performance improvements. • Results emphasize enhanced interpretability and generalizability for real-world clinical deployment. Deep learning models, particularly Convolutional Neural Networks (CNNs), have advanced automated Sleep Apnea (SA) detection from single-lead Electrocardiogram (ECG) signals. However, current approaches often face two limitations: (1) reliance on single-view representations (e.g., time-series or one image type) that can fail to capture the diverse feature set inherent in complex ECG dynamics; and (2) image-based methods that frequently employ standard architectures which may not optimally extract intricate local patterns crucial for SA identification. To address these challenges, we propose MVIC-Net, a Multi-View Interactive Convolutional Network framework that uniquely processes four complementary ECG representations simultaneously: 1D numeric data, 2D reshaped numeric data, Continuous Wavelet Transform (CWT) images, and circular plot images. We introduce the Convolutional and Interactive Learning Neural Network (CINet), an architecture inspired by SCINet and specifically engineered for enhanced ECG image feature extraction. CINet utilizes a hierarchical downsampling and interactive learning mechanism: it recursively splits feature maps, processes subsequences through distinct convolutional filters, and interactively combines them to mitigate information loss while capturing multi-resolution features effectively. Our MVIC-Net model integrates pre-trained CINet and standard CNN encoders, fusing their outputs via concatenation before final classification. Empirical validation on the Apnea-ECG database demonstrates superior performance, achieving 99.25% accuracy. This significantly surpasses individual-view models (with CINet reaching 96.32% accuracy on image-based views), pre-trained ResNet152V2 baselines, and a multi-model ensemble approach. To enhance clinical trust, we incorporate Explainable AI (XAI) via Grad-CAM, providing transparency into CINet’s decision-making process. Our results establish the effectiveness of combining multi-view learning with interactive convolutional architectures for robust physiological signal classification.

ABSTRACT The dynamic nature of Parkinson's disease (PD) is that it gradually impacts regions of the brain that are responsible for the production of the dopamine hormone. Despite continuous efforts, no effective treatment or preventative approach exists for PD. Nonetheless, the disease can be detected. Our goal is to create a Machine Learning and Deep Learning‐based system that can detect Parkinson's disease from a variety of data sources with high accuracy, sensitivity, specificity and interpretability. However, there have been significant advancements in the field of research, especially the use of artificial intelligence in the Parkinson's disease diagnostic process. We reviewed articles that were released between 2018 and 2024, concentrating on the most current studies that had been published. We chose 70 research articles for our review paper based on a set of criteria from a variety of online databases, including IEEExpress, medical databases like PubMed, Google Scholar, ResearchGate and ScienceDirect, and various publishers, including Elsevier, Taylor & Francis, Springer, MDPI, Plos One and so forth. According to our review, the majority of works make use of voice data. Our review study found that the highest accuracy level of most papers was above 90%, and the most commonly used algorithms were CNN and SVM. The main goal of this review study is to look into and put together information about the different ways that artificial intelligence, especially Machine Learning, can be used to find Parkinson's disease. Using diverse data gathered from multiple public and private datasets, we can infer that the application of artificial intelligence, particularly Machine Learning algorithms, for identifying Parkinson's disease plays a crucial role in the medical field.

The deployment of complex deep learning models in resource-limited clinical settings remains a critical challenge for brain tumor classification. This study explores knowledge distillation (KD) and ensembling strategies to enhance brain tumor classification using lightweight convolutional neural networks (CNNs), targeting resource-constrained environments. The research investigates both single-teacher and multi-teacher KD frameworks, utilizing VGG16 and Xception as teacher models to impart knowledge to smaller student models, including Tiny ResNet, Tiny Xception, and Tiny DenseNet. Student models, such as Student-1 (Tiny ResNet), feature a parameter reduction of approximately 83x and 117x times compared to VGG16 and Xception. Moreover, ensembling methods, including Average Ensembling and Weighted Ensembling, were employed to further refine predictive performance further. The multi-teacher KD approach achieved notable accuracy, reaching 96.19 % on Dataset-1 and 95.57 % on Dataset-2, while ensembling techniques yielded up to 98.02 % and 98.15 % for Average Ensembling and Weighted Ensembling, respectively. Furthermore, explainable AI (XAI) techniques, such as Grad-CAM and SHAP, were implemented to improve the interpretability of deep learning models and provide clinicians with insights. The results underscore the potential of combining KD and ensembling to develop robust and efficient AI-driven solutions for brain tumor classification, balancing accuracy, interpretability, and computational efficiency. These advancements make it possible to develop brain tumor classification systems that are accurate and efficient enough for deployment in resource-constrained environments.

Lung cancer, one of the most prevalent and deadliest diseases, necessitates early detection for patient survival. The low level of contrast between lesions and adjacent lung tissue, coupled with the diverse shapes and structures of lung nodules, poses significant challenges for their accurate identification and classification. Despite the extensive use of machine learning methods, the lack of adequately annotated datasets significantly hampers efficient model training. Moreover, the lack of transparency in deep learning models has led to their perception as “black boxes”, constraining their credibility for end users like radiologists. To address these issues, we present LungCT-NET , a novel transfer learning-based architecture coupled with ensemble learning and explainable AI for binary classification of lung nodules into malignant and benign using lung CT scans. LungCT-NET incorporates essential preprocessing, reconfiguring transfer learning models, and an advanced stacking ensemble strategy utilizing combinations of the top-performing pre-trained models. Several transfer learning algorithms are employed, including VGG-16, VGG-19, MobileNet-V2, InceptionNet-V3, EfficientNet-B0, ResNet152-V2, and DenseNet-121. Extensive experimental analyses have been carried out on the LIDC-DIRI dataset using various performance metrics for evaluation. The findings demonstrate that the suggested framework significantly exceeds state-of-the-art approaches, achieving an accuracy, precision, F1 score and recall of 98.99%, an AUC of 98.15%. Finally, the integrated SHapley Additive exPlanations (SHAP) enhance the grasp of model outcomes, hence increasing confidence in lung cancer prognosis. Therefore, the proposed innovative LungCT-NET can potentially support clinical settings by automating lung nodule classification from low-dose CT scans, aiding physicians and radiologists in prompt, accurate diagnoses.

Early diagnosis of brain tumors is critical for enhancing patient prognosis and treatment options, while accurate classification and segmentation of brain tumors are vital for developing personalized treatment strategies. Despite the widespread use of Magnetic Resonance Imaging (MRI) for brain examination and advances in AI-based detection methods, building an accurate and efficient model for detecting and categorizing tumors from MRI images remains a challenge. To address this problem, we proposed a deep Convolutional Neural Network (CNN)-based architecture for automatic brain image classification into four classes and a U-Net-based segmentation model. Using six benchmarked datasets, we tested the classification model and trained the segmentation model, enabling side-by-side comparison of the impact of segmentation on tumor classification in brain MRI images. We also evaluated two classification methods based on accuracy, recall, precision, and AUC. Our developed novel deep learning-based model for brain tumor classification and segmentation outperforms existing pre-trained models across all six datasets. The results demonstrate that our classification model achieved the highest accuracy of 98.7% in a merged dataset and 98.8% with the segmentation approach, with the highest classification accuracy reaching 97.7% among the four individual datasets. Thus, this novel framework could be applicable in clinics for the automatic identification and segmentation of brain tumors utilizing MRI scan input images.

Abstract Autism Spectrum Disorder (ASD) is a chronic condition characterised by impairments in social interaction and communication. Early detection of ASD is desired, and there exists a demand for the development of diagnostic aids to facilitate this. A lightweight Involutional Neural Network (INN) architecture has been developed to diagnose ASD. The model follows a simpler architectural design and has less number of parameters than the state‐of‐the‐art (SOTA) image classification models, requiring lower computational resources. The proposed model is trained to detect ASD from eye‐tracking scanpath (SP), heatmap (HM), and fixation map (FM) images. Monte Carlo Dropout has been applied to the model to perform an uncertainty analysis and ensure the effectiveness of the output provided by the proposed INN model. The model has been trained and evaluated using two publicly accessible datasets. From the experiment, it is seen that the model has achieved 98.12% accuracy, 96.83% accuracy, and 97.61% accuracy on SP, FM, and HM, respectively, which outperforms the current SOTA image classification models and other existing works conducted on this topic.

Museum guide robots should observe visitors in order to identify those who may desire a guide, and then initiate conversation with them. This paper presents a model for such robot behavior. Initiation of conversation is an important concern for social service robots such as museum guide robots. When people enter into a social interaction, they tend to situate themselves in a spatial-orientational arrangement such that each is facing inward around a space to which each has immediate access. When this kind of particular spatial formation occurs, they can feel that they are participating in the conversation; once they perceive their participation, they will subsequently try to maintain this spatial formation. We developed a model that describes the constraints and expected behaviors in initiation of conversation. Experimental results demonstrate that our model significantly improves a robot's performance in initiating conversation.

A mobile museum guide robot is expected to establish a proper spatial formation with the visitors. After observing the videotaped scenes of human guide-visitors interaction at actual museum galleries, we have developed a mobile robot that can guide multiple visitors inside the gallery from one exhibit to another. The mobile guide robot is capable of establishing spatial formation known as "F-formation" at the beginning of explanation. It can also use a systematic procedure known as "pause and restart" depending on the situation through which a framework of mutual orientation between the speaker (robot) and visitors is achieved. The effectiveness of our method has been confirmed through experiments.