Glioma, a primary malignant tumor of the central nervous system, presents significant challenges in intraoperative identification due to its infiltrative growth pattern. Traditional diagnostic methods, such as frozen sections and imprint cytology, are hindered by complex sample preparation and subjective interpretation. This study introduces an innovative system integrating droplet microfluidics and deep learning technologies to address cell overlap and unstable observation environments that have historically hindered accurate cell identification. Our optimized YOLOv8 architecture incorporates strategically placed attention mechanisms at layers 8, 9, and 15, enhancing feature extraction capabilities for small object detection while maintaining computational efficiency. The system’s robust multifocal analysis capability enables accurate distinction between glioma cells and normal cells across different focal planes, surpassing current imaging-based diagnostic methods. Through validation against conventional fluorescence detection methods, this work demonstrates superior stability in cell counting without threshold dependence and real-time processing with minimal latency. This integrated approach advances intraoperative diagnostics with a practical solution that can improve precision in glioma surgery and patient outcomes.