<![CDATA[Manual detection of pneumonia from X-ray images still faces challenges due to the long processing time, high cost, and strong dependence on radiologist expertise. This dependence increases the risk of delayed diagnosis and interpretation errors, potentially worsening patient conditions. To address these issues, this study proposes optimizing pneumonia detection using deep learning through the application of Binary Statistical Image Feature (BSIF) feature extraction. BSIF highlights important texture patterns in X-ray images to enhance the model’s ability to recognize pneumonia affected lung areas. The dataset consists of 2,239 chest X-ray images divided into two categories: normal lungs and pneumonia. The research stages include image preprocessing, BSIF feature extraction, model training using Convolutional Neural Network (CNN) and Vision Transformer (ViT) architectures, and performance evaluation based on precision, recall, f1-score, specificity, and ROC AUC. The results show that the CNN+BSIF combination achieved the best performance with 99.69% training accuracy and 79.17% validation accuracy, precision 87%, recall 72%, f1-score 74%, specificity 45.30%, and ROC AUC 94.08%. Meanwhile, ViT+BSIF reached 99.35% accuracy, CNN without BSIF 98.24%, and ViT without BSIF 90.16%. Therefore, CNN+BSIF proved to be the most optimal method for fast and accurate pneumonia detection.]]>
Copyrights © 2025
