20 June 2026 in Journal of Artificial Intelligence and Technology
Pneumonia remains a significant health challenge for children, particularly in regions with limited diagnostic resources. Early and accurate detection is essential for timely treatment and improved clinical outcomes. This paper presents an automated deep learning (DL)-based framework for pediatric pneumonia detection (PPD) using chest X-ray images, designed to support clinical decision-making and reduce diagnostic workload. Five convolutional neural networks (CNNs) architectures: DenseNet121, MobileNetV2, VGG16, InceptionV3, and ResNet50 are trained and evaluated through extensive experimentation. Their performance is assessed using accuracy, precision, recall, and F1-score, alongside confusion matrices. Results demonstrate that DenseNet121 and MobileNetV2 achieved superior performance, with an accuracy of 93.75%, F1-scores of nearly 95.1 %, outperforming the other models in balancing sensitivity and specificity. VGG16 achieves competitive performance with high sensitivity, whereas InceptionV3 and ResNet50 exhibit limitations, particularly in terms of generalization and specificity. The system also demonstrates scalability potential, with MobileNetV2 offering lightweight deployment capabilities for low-resource environments. The findings confirm the clinical value of DL systems in supporting PPD and highlight the promise of DenseNet121 and MobileNetV2 as practical solutions for real-world healthcare applications.
1 February 2026 in Bulletin of Electrical Engineering and Informatics
The caesarean section is one of the most frequently performed surgical procedures worldwide, with profound implications for maternal and neonatal health. Accurate prediction of delivery mode is essential for guiding clinical decisions, minimizing unnecessary surgical interventions, and improving patient outcomes. This study introduces a deep neural learning technique based on a temporal convolutional neural network (DNLTC) to classify delivery type—caesarean section versus normal vaginal delivery using maternal and obstetric data. The proposed model was evaluated against traditional machine learning (ML) approaches, including artificial neural networks (ANN), support vector machines (SVM), and decision trees (DT). Experimental results show that the DNLTC achieved the highest overall accuracy (85%), surpassing ANN (80%), SVM (68.8%), and DT (65%). TCNN also demonstrated strong clinical reliability, with a sensitivity of 94%, specificity of 91%, and a perfect F1-score of 100%. These findings highlight the advantages of incorporating temporal feature learning into delivery mode prediction, enabling the detection of subtle, sequential patterns that conventional models may overlook. By providing more accurate and robust predictions, the proposed framework can support obstetricians in making timely, evidence-based decisions, ultimately enhancing maternal and newborn health outcomes.