DOI: 10.46810/tdfd.1871343 ISSN: 2149-6366

A DEEP LEARNING APPROACH BASED ON YOLOV11M FOR CLASSIFYING COVID-19 AND PNEUMONIA ON CHEST X-RAY IMAGES

Berin Balcı, Fatih Basciftci
Show PDF Cite
COVID-19 and pneumonia are among the most common respiratory diseases worldwide and are linked to high morbidity and mortality. Rapid and reliable detection is essential for effective clinical management. This study applies the YOLOv11m deep learning model to classify chest X-ray images into three categories: COVID-19, Normal, and Pneumonia. The dataset, obtained from the publicly available Kaggle repository, includes 1,518 radiographs (COVID-19: 363; Normal: 1,020; Pneumonia: 135) and is divided into training (80%), testing (10%), and validation (10%) subsets. A three-stage preprocessing pipeline was used to isolate pulmonary regions: Contrast-Limited Adaptive Histogram Equalization (CLAHE) for contrast enhancement, Otsu thresholding with morphological operations for binary lung mask generation, and cropping of lung areas with 3% padding before resizing to 224 × 224 pixels. The model was trained using the PyTorch framework. Performance evaluation included precision, recall, F1-score, ROC-AUC, and Top-1/Top-5 accuracy. YOLOv11m achieved 98.16% Top-1 accuracy, 100% Top-5 accuracy, a macro F1-score of 0.9708, and a macro ROC-AUC of 0.9993. Class-specific F1-scores were 98.0% for COVID-19, 98.7% for Normal, and 94.5% for Pneumonia. These findings demonstrate that YOLOv11m can effectively distinguish COVID-19, Normal, and Pneumonia patterns on chest radiographs, highlighting its strong potential for AI-assisted diagnostic and clinical decision-support applications.

More from our Archive