The rapid advancement of medical imaging technologies requires the development of advanced, automated, and interpretable diagnostic tools for clinical decision-making. Although convolutional neural networks (CNNs) have shown significant promise in medical image analysis, they have limitations in capturing the global context and lack interpretability, thereby hindering their clinical adoption. This study presents EFFResNet-ViT, a novel hybrid deep learning (DL) model designed to address these challenges by combining EfficientNet-B0 and ResNet-50 CNN backbones with a vision transformer (ViT) module. The proposed architecture employs a feature fusion strategy to integrate the local feature extraction strengths of CNNs with the global dependency modeling capabilities of transformers. The extracted features are further refined through a post-transformer CNN and a global average pooling layer to enhance the classification performance. To improve interpretability, EFFResNet-ViT incorporates Grad-CAM visualization techniques to highlight regions contributing to classification decisions and employs t-distributed stochastic neighbor embedding for feature space analysis, providing insights into class separability. The proposed model was evaluated on two benchmark datasets: brain tumor (BT) CE-MRI for brain tumor classification and a retinal image dataset for ophthalmological diagnosis. EFFResNet-ViT achieved state-of-the-art performance, with accuracies of 99.31% and 92.54% on the BT CE-MRI and retinal datasets, respectively. Comparative analyses demonstrate the superior classification performance and interpretability of EFFResNet-ViT over existingViT and CNN-based hybrid models. The explainable design of EFFResNet-ViT addresses the critical need for transparency in artificial intelligence-driven medical diagnostics, facilitating its potential integration into clinical workflows to improve decision-making and patient outcomes.

Precise segmentation is vital for successful diagnosis and treatment planning. Medical image segmentation has demonstrated remarkable advances with the introduction of deep convolutional neural networks, particularly encoder-decoder networks such as U-Net. Despite their excellent performances, these methods have some limitations. First, the structure is limited in its ability to combine information because feature maps to extract valid information from the final encoding stage are incompatible at the encoding and decoding levels. Second, the approach ignores significant semantic details and does not consider different types of small-scale contextual information when segmenting medical images. Lastly, most methods employing 3D architectures to process input medical images increase the computational complexity of the model without significantly improving the accuracy. To resolve these issues, we propose a segmentation network called Multi-Attention Gated Residual U-Net (MAGRes-UNet). This network incorporates four multi-attention gate (MAG) modules and residual blocks into a standard U-Net structure. The MAG module integrates the information from all encoding stages and focuses on small-scale tumors while disambiguating irrelevant and noisy feature responses, thereby promoting meaningful contextual information. The residual blocks simplify the network training and mitigate the problem of vanishing gradients. This improves the ability of the network to effectively learn intricate features and deep representations. Moreover, our network employs the Mish and ReLU activation functions (AFs), which utilize AdamW and Adam optimization strategies to achieve enhanced segmentation performance. The proposed MAGRes-UNet method was compared with the U-Net, Multi-Attention Gated-UNet (MAG-UNet), and Residual-UNet (ResUNet) models. In addition, a statistical T-test was performed to assess the difference in model significance between the approaches. The analysis revealed that MAGRes-UNet employing Mish and AdamW provides significant performance improvement over the ReLU AF and Adam optimizer on two benchmark datasets: Multi-Class BT T1-weighted Contrast-Enhanced Magnetic Resonance Imaging (T1-CE-MRI) and skin lesions HAM10000 (Human Against Machine with 10,000 training images). MAGRes-UNet using Mish and AdamW provides competitive performance over the representative medical image segmentation methods.

Brain tumors (BT) present a considerable global health concern because of their high mortality rates across diverse age groups. A delay in diagnosing BT can lead to death. Therefore, a timely and accurate diagnosis through magnetic resonance imaging (MRI) is crucial. A radiologist makes the final decision to identify the tumor through MRI. However, manual assessments are flawed, time-consuming, and rely on experienced radiologists or neurologists to identify and diagnose a BT. Computer-aided classification models often lack performance and explainability for clinical translation, particularly in neuroscience research, resulting in physicians perceiving the model results as inadequate due to the black box model. Explainable deep learning (XDL) can advance neuroscientific research and healthcare tasks. To enhance the explainability of deep learning (DL) and provide diagnostic support, we propose a new classification and localization model, combining existing methods to enhance the explainability of DL and provide diagnostic support. We adopt a pre-trained visual geometry group (pre-trained-VGG-19), scratch-VGG-19, and EfficientNet model that runs a modified form of the class activation mapping (CAM), gradient-weighted class activation mapping (Grad-CAM) and Grad-CAM++ algorithms. These algorithms, introduced into a convolutional neural network (CNN), uncover a crucial part of the classification and can provide an explanatory interface for diagnosing BT. The experimental results demonstrate that the pre-trained-VGG-19 with Grad-CAM provides better classification and visualization results than the scratch-VGG-19, EfficientNet, and cutting-edge DL techniques regarding visual and quantitative evaluations with increased accuracy. The proposed approach may contribute to reducing the diagnostic uncertainty and validating BT classification.

Enhancing cybersecurity in DC shipboard microgrid (SMG) systems is crucial for maintaining the resilience of energy operation in maritime transportation systems (MTS). However, increasing cyber threats pose significant challenges to deploying resilient technologies in intelligent DC SMGs. These intricate cyber-physical systems, comprising power electronics, distributed generation units, sensors, and monitoring technologies, are managed remotely, making them vulnerable to attacks that jeopardize their stability and security. Existing solutions often require additional support due to low detection and high false alarm rates, primarily from manual analysis. To address these issues, this study presents a sophisticated deep learning-driven cyber-attack detection and identification model designed to effectively enhance the security of DC SMGs in MTS. The proposed framework leverages Long Short-Term Memory (LSTM) with variational autoencoder (VAE) architectures for deep feature extraction (DFE) under attack scenarios. VAE schemes automatically uncover hidden patterns within the DC SMG network, and their outputs are utilized by deep learning (DL) schemes for accurate attack detection. The proposed model incorporates a deep artificial neural network (ANN)-based encoder-decoder scheme to identify attacks in DC SMGs, contributing to an efficient operation. Additionally, DL-based LSTM-VAE and ANN are meticulously designed with appropriate hyperparameters to counter cyber threats. The data-driven DL method achieved the testing accuracy of 99.81%, outperforming various state-of-the-art (SOTA) DL and machine learning (ML) methods. Extensive testing scenarios have been conducted to demonstrate the performance and robustness of the proposed DL methods under different levels of attacks, ensuring their efficacy in enhancing the cybersecurity of DC shipboard microgrids.