<![CDATA[Pneumonia is a lung infection disease that remains one of the leading causes of death worldwide, particularly among children and vulnerable populations. The diagnosis process using Chest X-Ray (CXR) images still faces several challenges, including the limited number of radiologists and the potential for human error in medical image interpretation. This study aims to perform a comparative analysis of YOLOv5, YOLOv8, and YOLOv9 small and medium variants for bounding box-based pneumonia detection on chest X-Ray images. The dataset was obtained from Kaggle and processed using Roboflow through preprocessing stages including auto-orient and image resizing to 640×640 pixels. The dataset was divided into 70% training, 20% validation, and 10% testing data. The training process was conducted using the Ultralytics YOLO framework for 50 epochs on an NVIDIA Tesla T4 GPU in Google Colaboratory. Model evaluation was carried out using Precision, Recall (Sensitivity), mAP50, and mAP50-95 metrics. The testing results showed that the YOLOv9m model achieved the best performance with a Recall value of 0.982759 and an mAP50-95 value of 0.717609. In addition, YOLOv5m produced the highest Precision value of 0.948602 and mAP50 of 0.969271. Based on the experimental results, the YOLOv9 architecture demonstrated the most optimal performance for pneumonia detection on Chest X-Ray images compared to the other models. ]]>
Copyrights © 2026
