Real-time detection of circulating tumor cells (CTCs) and cancer associated fibroblasts (CAFs) in microscopic images is crucial for determining prognosis, monitoring disease progression, and evaluating treatment effectiveness. While there have been significant strides in deep learning-based While there have been significant strides in deep learning-based cell detection methods, accurately identifying target cells remains a challenging task due to the high density of CTCs and CAFs at intersections in a large number of images. Our study addresses this issue by proposing CYOLOv8, a real-time method for detecting CTCs and CAFs improved YOLOv8. Firstly, the YOLOv8 backbone network was used for the detection of CTCs and CAFs. YOLOv8 backbone network's ability to extract both local and global features from cellular images is enhanced by adding the BOT module, which combines CNN and Transformer benefits. Additionally, by replacing the C2f module with the C2f-SCConv module, the YOLOv8 neck network can maintain its effectiveness. Finally, the EMA attention module is incorporated after the BOT module and the C2f-SCConv module. the C2f-SCConv module, thereby ensnaring the network to focus more on the target feature information and eliminate the hindrance of extraneous information. By conducting experiments on the CAC dataset, the mAP50 and mAP50-95 of CYOLOv8 on the validation set reached an impressive 97.3% and 79.8%, respectively. These results are an improvement of 1.1 and 4.2 percentage points compared to the unaltered YOLOv8 model while reducing the number of parameters and computation by 6% and 3.7%, respectively. The method's validity is confirmed by comparing it with other well-known techniques, indicating that it is an effective means of detecting CTC and CAF in medical images