Wyniki wyszukiwania - BazTech

Ograniczanie wyników

1 Polish Journal of Medical Physics and Engineering

1 2025

Znaleziono wyników: 1

Liczba wyników na stronie

Wyniki wyszukiwania

Sortuj według:

Ogranicz wyniki do:

Lung diseases classification using pre-trained based deep learning model and support vector machine

Slama Amine Ben, Amri Yessine, Barbaria Sabri, Rahmouni Hanene Boussi, Trabelsi Hedi

Polish Journal of Medical Physics and Engineering

2025

Vol. 31, Iss. 3

178--194

Introduction: Given its rapid transmission and heightened fatality rate, early detection of viral pneumonia is imperative. The virus mainly affects the lung, leading to pneumonia alongside symptoms like fatigue, dry cough, and fever, which can sometimes be misdiagnosed as other respiratory conditions such as lung cancer or pneumonia. Chest X-rays are widely used in the healthcare sector to provide both swift and accurate diagnoses. Deep learning algorithms have demonstrated significant efficiency in detecting and classifying lung diseases, which enhanced the diagnostic process and saving valuable time for medical applications and therapy.The objective of this study is to develop and evaluate a deep learning-based architecture for the accurate multi-class classification of respiratory diseases, including pneumonia, lung opacity, and COVID-19, using chest X-ray images to enhance diagnostic efficiency in healthcare settings. Material and methods: A substantial dataset comprising X-ray images was crated, including 1026 pneumonia cases, 1256 COVID-19 cases, 2305 lung opacity cases and 3224 normal X-ray images. For classification purposes, we employed a pre-trained VGG19 model combined with an SVM classifier. To validate the model’s accuracy, we utilized cross-validation techniques and performance metrics, including precision, recall, F1-score, and the area under the curve (AUC). This approach ensures robust evaluation of the proposed framework. Results: The experimental results demonstrated the superiority of our proposed VGG16+SVM model over existing approaches, achieving an accuracy of 93.50%, recall of 94.16%, precision of 94.45%, F1 score of 93.28%, and area under the curve (AUC) of 90.16%. Conclusions: This enhanced performance equips healthcare practitioners with the tools to diagnose and treat patients more expeditiously and effectively.