Background: Skin cancer, particularly melanoma, poses significant challenges due to the heterogeneity of skin images and the demand for accurate and interpretable diagnostic systems. Early detection and effective management are crucial for improving patient outcomes. Traditional AI models often struggle with balancing accuracy and interpretability, which are critical for clinical adoption. Methods: The SmartSkin-XAI methodology incorporates a fine-tuned DenseNet121 model combined with XAI techniques to interpret predictions. This approach improves early detection and patient management by offering a transparent decision-making process. The model was evaluated using two datasets: the ISIC dataset and the Kaggle dataset. Performance metrics such as classification accuracy, precision, recall, and F1 score were compared against benchmark models, including DenseNet121, InceptionV3, and esNet50. Results: SmartSkin-XAI achieved a classification accuracy of 97% on the ISIC dataset and 98% on the Kaggle dataset. The model demonstrated high stability in precision, recall, and F1 score measures, outperforming the benchmark models. These results underscore the robustness and applicability of SmartSkin-XAI for real-world healthcare scenarios. Conclusions: SmartSkin-XAI addresses critical challenges in melanoma diagnosis by integrating state-of-the-art architecture with XAI methods, providing both accuracy and interpretability. This approach enhances clinical decision-making, fosters trust among healthcare professionals, and represents a significant advancement in incorporating AI-driven diagnostics into medicine, particularly for bedside applications.

Sustainable agriculture holds the key in meeting food production requirements for a rapidly growing population without exacerbating environmental degradation. Plant leaf diseases pose a critical threat to crop yield and quality. Existing inspection methods are labor‐intensive and prone to human errors, while lacking support for large‐scale agriculture. This research aims to enhance plant health by developing advanced deep learning models for the detection and classification of plant diseases across a variety of species. A deep learning model based on the paradigm of the MobileNet architecture is proposed, which employs a dedicated design through deeper convolutional layers, dropout regularization, and fully connected layers. This results in significant improvements in disease classification in tomato, bean, and chili plants, with accuracy rates of 97.90%, 98.12%, and 97.95%, respectively. Moreover, Grad‐CAM is used to shed light on the decision‐making process of the proposed model. The work contributes to the advancement of precision farming and sustainable agricultural practices, supporting timely and accurate plant disease diagnosis.

Alzheimer's disease (AD), the most prevalent cause of dementia, affects over 55 million individuals globally. With aging populations, AD cases are expected to increase substantially, presenting a pressing public health challenge. Early diagnosis is crucial but remains challenging, particularly in the mild cognitive impairment stage before extensive neurodegeneration. Existing diagnostic methods often fall short due to the subtle nature of early AD symptoms, highlighting the need for more accurate and efficient approaches. In response to this challenge, we introduce a hybrid framework to enhance the diagnosis of Alzheimer's Disease (AD) across four classes by integrating various deep learning (DL) and machine learning (ML) techniques on an MRI image dataset. We applied multiple preprocessing techniques to the MRI images. Then, the methodology employs three pre-trained convolutional neural networks (CNNs): VGG-16, VGG-19, and MobileNet - each undergoing training under diverse parameter settings through transfer learning to facilitate the extraction of meaningful features from images, utilizing convolution and pooling layers. Subsequently, for feature selection, a decision tree-based RFE method was employed to iteratively select the most significant features and enable more accurate AD classification. Finally, an XGBoost classifier was used to classify the multiclass types of AD under 5-fold cross-validation to assess the performance of our proposed model. The proposed model achieved the highest accuracy of 93% for multiclass classification, indicating that our approach significantly outperforms state-of-the-art methods. This model could apply to clinical applications, marking a significant advancement in AD diagnostics.