2026/04/22 by Z. Bager Dahoos, S. Elbai, M. A. Alghrawy +4 · 1 voice
Computer Science · Medicine · #AI in cancer detection #COVID-19 diagnosis using AI #Phonocardiography and Auscultation Techniques
paper · pdf · doi:10.26437/9gzkq811
openalex created_date 2026/04/22 · openalex publication_date 2026/04/22 · openalex updated_date 2026/06/11
Purpose: The study aims to perfect the accuracy and reliability of pneumonia detection in chest X-ray pictures, as pneumonia remains a major cause of respiratory-related deaths, particularly among children and the elderly, while manual diagnosis is often time-consuming and prone to variability and errors. Design/Methodology/Approach: The study proposes an integrated hybrid detection framework that combines unsupervised clustering techniques such as K-means with traditional deep learning models, where deep neural networks extract high-level features, and clustering identifies hidden data patterns, and the outputs are integrated using soft voting and majority voting techniques, with hyperparameter tuning applied to reduce overfitting and enhance performance. Findings: The proposed FBMV model achieved superior results, reaching 100% training accuracy, 99.51% validation accuracy, and 99.38% precision with the lowest loss values, while MobileNet demonstrated the fastest execution time suitable for real-time applications, and VGG16 showed the lowest memory usage, making it suitable for well-constrained environments. Research Limitation: The study is limited by its reliance on a specific high-quality chest X-ray dataset, which may not fully represent real-world variability, and further validation is required across diverse clinical settings and populations. Practical Implication: The proposed model can support healthcare professionals in achieving faster, more accurate pneumonia diagnoses, reducing errors and improving clinical decision-making, especially in limited-resource environments. Social Implication: Enhancing pneumonia detection contributes to early treatment, reduced mortality rates, and improved healthcare outcomes, particularly for vulnerable groups such as children and the elderly. Originality/Value: The study presents a novel hybrid range that integrates unsupervised clustering with deep learning and ensemble voting techniques (FBMV), providing a more stable and accurate diagnostic approach and adding value through the efficiency of MobileNet and the low-resource capabilities of VGG16.