Introduction: Diabetic retinopathy (DR) is an inflammatory condition affecting the retina caused by elevated and unregulated blood glucose levels. On a global scale, it is a contributing factor to vision impairment. Several deep learning (DL) methods use retinal images to identify DR severity. However, a significant improvement is required to assist medical professionals in recognizing DR in its early phases. Methods: Thus, the author introduced a method based on the DL technique to determine the DR severity grades using retinal images. A ShuffleNet V2 model with vision transformers' (ViT) attention mechanism was used to extract the features. An improved Whale optimization method (IWO) was used to fine-tune the feature extraction model. We employed a convolutional Kolmogorov-Arnold Network to categorize the DR severity using the extracted features. The EyePACS dataset was utilized to train the proposed DR severity grading model using a five-fold cross-validation strategy. We generalized the model on the Messidor-2 dataset. Results: The findings revealed an average accuracy of 93.84% on the MESSIDOR-2 dataset, demonstrating a substantial improvement in detecting DR using the fundus images. Discussion: Furthermore, the model demands minimal processing resources to generate the outcomes, leading to the deployment of the proposed DR severity detection model in healthcare facilities with limited computational resources.
Down syndrome (DS) is one of the prevalent chromosomal disorders, representing distinctive craniofacial features and a range of developmental and medical challenges. Due to the lack of clinical expertise and high infrastructure costs, access to genetic testing is restricted to resource-constrained clinical settings. There is a demand for developing a non-invasive and equitable DS screening tool, facilitating DS diagnosis for a wide range of populations. In this study, we develop and validate a robust, interpretable deep learning model for the early detection of DS using facial images of infants. A hybrid feature extraction architecture combining RegNet X-MobileNet V3 and vision transformer (ViT)-Linformer is developed for effective feature representation. We use an adaptive attention-based feature fusion to enhance the proposed model's focus on diagnostically relevant facial regions. Bayesian optimization with hyperband (BOHB) fine-tuned extremely randomized trees (ExtraTrees) is employed to classify the features. To ensure the model's generalizability, stratified five-fold cross-validation is performed. Compared to the recent DS classification approaches, the proposed model demonstrates outstanding performance, achieving an accuracy of 99.10%, precision of 98.80%, recall of 98.87%, F1-score of 98.83%, and specificity of 98.81%, on the unseen data. The findings underscore the strengths of the proposed model as a reliable screening tool to identify DS in the early stages using the facial images. This study paves the foundation to build equitable, scalable, and trustworthy digital solution for effective pediatric care across the globe.
face imagesinfantsnon-invasive screeninghybrid feature extraction
Background As a crucial cognitive and motor function, handwriting can reveal dyslexia, dysgraphia, and other neurodevelopmental conditions. With the emergence of AI, deep learning (DL) models can analyze handwriting patterns for early identification and categorization of these conditions with high level of accuracy. Objective This study presents a literature review by examining DL models in classifying learning disabilities using handwriting data. It identifies models’ shortcomings and suggests future research directions, improving model accuracy, fairness, and usability. Methodology By employing the standard literature guidelines, the authors identified 24 studies from a pool of 502 studies. They extracted dataset details, methodologies, and key findings from the studies. Findings The findings indicate the dominance of convolutional neural networks in improving learning disabilities classification through the extraction of fine-grained spatial handwritten patterns. Vision transformers (ViTs) show potential for long-range contextual feature extraction to enhance the model’s generalization capabilities. Despite these advances, dataset size, model generalizability, computational efficiency, interpretability, and real-world deployment continue to be significant challenges. Conclusions The review findings underscore the transformative potential of DL-based handwriting analysis in educational and clinical settings, providing early detection and intervention in learning disabilities.
Across the globe, chronic obstructive pulmonary disease (COPD) continues to be a critical healthcare burden due to its progressive and undetected decline in pulmonary function. In recent times, convolutional neural networks (CNNs) and vision transformers (ViTs) have been widely applied to lung sounds (LS) analysis, leading to the development of improved non-invasive, accessible, and scalable COPD screening methods. However, the shortcomings of CNNs and ViTs architectures restrict the COPD classifier’s discriminative power in distinguishing normal and COPD samples. In this study, a unique and interpretable COPD classification is proposed. This study introduces a hybrid feature extraction strategy integrating CNNs and ViTs architecture to capture the crucial COPD patterns from the LS’s Mel-spectrogram representation. The proposed CNNs-ViTs architecture addresses the limitations of standalone CNNs or ViTs architectures by capturing fine-grained acoustic features and long-range temporal dependencies while efficiently modeling non-linear relationships. A novel feature fusion is used to combine the heterogeneous features into a unified and highly discriminative feature representation space. A tailored hyperparameter optimization strategy is developed to fine-tune the Kolmogorov-Arnold Networks (KANs) for classifying the features into normal and COPD classes. By incorporating internal (552 Healthy and 1324 COPD LS recordings) and external (35 Healthy and 77 COPD LS recordings) datasets, the authors evaluate the model’s performance. The model achieves a generalization accuracy of 97.62% and specificity of 96.51% on the external dataset. Through Shapley Additive Explanations (SHAP) overlays, the model’s interpretability is enhanced. The experimental findings highlight the effectiveness of the proposed COPD classification architecture in the domain of biomedical acoustic signal processing. Operating with minimal reliance on specialized diagnostic infrastructure renders the proposed model ideal for primary care and resource-limited environments.
Microscopy Research and Technique · مجلد 89 (5) · 687-704
Prostate cancer is a prevalent and serious health concern, ranking among the most frequently diagnosed cancers and a leading cause of cancer-related deaths in men worldwide. Early detection and accurate diagnosis are crucial for improving patient outcomes by limiting disease progression. Histopathological image analysis remains the gold standard for prostate cancer detection; however, manual interpretation is time-consuming and requires specialized expertise. To address these challenges, this study proposes a hybrid deep learning framework that combines an ensemble of transfer-learned CNNs (VGG-16, DenseNet-121, and AlexNet) with a fine-tuned Vision Transformer (ViT). The CNN ensemble extracts rich local features, while the ViT captures global contextual dependencies through a self-attention mechanism and a multilayer perceptron (MLP). Additionally, a cross-attention fusion (CAF) module integrates local and global features, and knowledge distillation (KD) enables a lightweight student network suitable for efficient clinical deployment. The study utilizes the publicly available PANDA dataset for training and testing. Preprocessing steps, including patch generation, gamma correction, and stain deconvolution, enhance image quality and feature representation. A comprehensive evaluation was conducted using standard performance metrics such as accuracy, true positive rate (TPR), true negative rate (TNR), precision, F1-score, false negative rate (FNR), and false positive rate (FPR). An ablation study confirmed the contribution of each module, highlighting the critical role of ensemble CNNs, CAF, and ViT in improving performance. Experimental results demonstrate that the proposed model outperforms conventional transfer learning models and existing state-of-the-art techniques, achieving 97.91% accuracy, along with significant improvements in TPR, TNR, and reduced FNR/FPR. The computational complexity, evaluated in terms of parameters, FLOPs, GPU memory, and inference time, indicates that the proposed model is more demanding than traditional CNNs. Nevertheless, the architecture strikes a practical balance between predictive accuracy and efficiency, making it suitable for real-world clinical applications. These findings underscore the potential of AI-powered hybrid models in expediting prostate cancer diagnosis and enabling timely intervention for improved patient outcomes.
prostate cancer detectionhistopathology image analysishybrid CNN-ViT modelcross-attention fusion
Advanced Health Science and Technology Journal · مجلد 1 (2) · 92-99
Abstract Background: Epilepsy remains one of the most prevalent neurological disorders worldwide, characterized by recurrent seizures resulting from abnormal electrical brain activity. Traditional diagnostic methods, such as electroencephalography (EEG) and magnetic resonance imaging (MRI), although widely used, often fail to provide sufficient accuracy when applied in isolation due to their inherent limitations in temporal or spatial resolution. These challenges highlight the need for integrated analytical approaches capable of leveraging complementary modalities for improved diagnostic precision. Objectives: This study aims to address the limitations of single-modality diagnostic techniques by proposing a multimodal deep learning framework that integrates EEG and MRI data to enhance the accuracy of epilepsy detection. The objectives include preprocessing EEG signals to remove artifacts, segmenting MRI scans for structural feature extraction, employing CNNs to model EEG spatio-temporal dependencies, using ViTs to capture hierarchical MRI features, and integrating these representations through a cross-attention fusion mechanism for effective supervised classification. Method: The methodology involves preprocessing EEG signals to eliminate artifacts and ensure clean temporal representation, along with MRI scan segmentation to extract structural information. Convolutional neural networks (CNNs) were used to model spatio-temporal patterns within EEG signals, while vision transformers (ViTs) extracted spatially hierarchical features from MRI data. A cross-attention fusion mechanism was designed to integrate these complementary modalities, and a supervised classifier performed the final prediction. Model training and evaluation were carried out using a high-performance computing setup with GPU acceleration to efficiently handle multimodal data. Results: Experimental results demonstrated that the proposed multimodal deep learning model significantly outperformed state-of-the-art pretrained CNN and ViT baselines across major evaluation metrics, including accuracy, precision, recall, F1-score, and Matthews correlation coefficient. The confusion matrix further confirmed robust differentiation between normal and epileptic classes, emphasizing the reliability of the proposed framework for accurate clinical decision-making. Conclusion: The findings of this study indicate that integrating EEG and MRI data through a multimodal fusion framework provides a substantial improvement in epilepsy detection compared to single-modality approaches. The results highlight the importance of combining complementary temporal and structural information to achieve reliable diagnostic performance. This work contributes to advancing early detection, reducing diagnostic errors, and supporting timely treatment strategies, reinforcing the value of multimodal deep learning approaches in personalized healthcare.
Abstract The timely diagnosis and identification of apple leaf diseases is essential to prevent the spread of diseases and ensure the sound development of the apple industry. Convolutional neural networks (CNNs) have achieved phenomenal success in the area of leaf disease detection, which can greatly benefit the agriculture industry. However, their large size and intricate design continue to pose a challenge when it comes to deploying these models on lightweight devices. Although several successful models (e.g., EfficientNets and MobileNets) have been designed to adapt to resource-constrained devices, these models have not been able to achieve significant results in leaf disease detection tasks and leave a performance gap behind. This research gap has motivated us to develop an apple leaf disease detection model that can not only be deployed on lightweight devices but also outperform existing models. In this work, we propose AppViT, a hybrid vision model, combining the features of convolution blocks and multi-head self-attention, to compete with the best-performing models. Specifically, we begin by introducing the convolution blocks that narrow down the size of the feature maps and help the model encode local features progressively. Then, we stack ViT blocks in combination with convolution blocks, allowing the network to capture non-local dependencies and spatial patterns. Embodied with these designs and a hierarchical structure, AppViT demonstrates excellent performance in apple leaf disease detection tasks. Specifically, it achieves 96.38% precision on Plant Pathology 2021—FGVC8 with about 1.3 million parameters, which is 11.3% and 4.3% more accurate than ResNet-50 and EfficientNet-B3. The precision, recall and F score of our proposed model on Plant Pathology 2021—FGVC8 are 0.967, 0.959, and 0.963 respectively.
apple leaf disease detectionlightweight vision transformerplant disease classificationmulti-head self-attention