<![CDATA[Bloodstream infection accounts for approximately 11 million deaths annually, and yet conventional blood culture methods require 40-48 hours to complete pathogen identification which delays definitive therapeutic decisions. Gram staining does provide preliminary bacterial classification within hours, but manual interpretation still remains a labor-intensive task and is prone to variability. This study develops an automated bacterial detection and classification system by integrating CBAM into the YOLOv8 architecture. The model was trained on Gram-stained microscopic images across four bacterial categories: Gram-positive cocci, Gram-negative cocci, Gram-positive bacilli, and Gram-negative bacilli. Dataset preprocessing involved quality selection, noise reduction, and targeted augmentation to address severe class imbalances. The inclusion of CBAM improved feature discrimination and localization performance, with an increase of 1.4% in mAP@0.5:0.95 (from 70.8% to 72.2%). The proposed model also reduced cross-class misclassifications, particularly among morphologically similar cocci. These findings demonstrate that integrating lightweight attention mechanisms can enhance bacterial detection reliability in microscopic imaging and support the development of automated systems for faster, more consistent preliminary bacterial identification.]]>
Copyrights © 2025
