Exploring Two-Stage Multi-Class Pneumonia Prediction from x-ray Images with Lightweight Feature Extraction Model

Pneumonia is an acute respiratory infection of one or both lungs caused by bacteria, viruses, or fungal agents. Chest radiography is widely used to diagnose pneumonia because it exhibits abnormal air gaps or partially solid areas, indicating a blockage through which no air can pass. The proper classification of pneumonia from chest x-ray images is critical for prompt diagnosis and treatment. This research employs an enhanced pre-trained transfer learning model to conduct multi-class pneumonia prediction with excellent classification precision. The suggested methodology begins with pre-processing strategies to improve image quality and mitigate noise, followed by advanced feature extraction algorithms to extract discriminative details from the images. The lightweight EfficientNet model serves as the foundation for feature extraction, leveraging its learned representations through two-stage training while fine-tuning the model for the specific objective of distinguishing between pneumonia and COVID-19 cases. The model’s main benefit is its ability to strike an acceptable balance between accuracy and processing resources, making it suitable for this application. Extensive assessments were carried out utilizing radiography chest x-ray images, and the suggested algorithm was compared to other state-of-the-art approaches. The experimental findings reveal that the suggested strategy outperforms alternative approaches in qualitative analysis, indicating its potential for clinical applications.