Algorithm Improvement for Bacterial Detection in Microfluidic Environments

In this work, we applied a combination of microfluidic chip technology and microscopic imaging to observe bacterial growth under antibiotic influence allows for faster and more accurate assessment of antibiotic efficacy against bacteria, which is crucial for clinical diagnosis and prevention of antibiotic misuse. This study proposes an algorithm improvement method for E. coli detection on a microfluidic platform based on YOLOv8, which effectively addressed issues such as missed detections in small target detection like bacteria. Firstly, a dense connection mechanism is introduced in the C2f module. Secondly, Attention-based Intra-scale Feature Interaction is added deep within the backbone to enhance the model's capability in recognizing and localizing small targets. Additionally, an iterative attention-based multi-scale feature fusion method is incorporated into the Neck, which not only reduces computational costs but also allows the model to focus more on relevant features, minimizing interference from redundant information. In image preprocessing, we applied contrast limited adaptive histogram equalization to the captured bacterial images. According to experimental results, the model has improved mAP by approximately 10% after preprocessing the bacterial images. Compared to the Baseline, the improved model achieved a slight decrease in accuracy but increased recall and [email protected] by 5.4% and 2.4%, respectively.