Objective To explore an automated, rapid, and objective method for measuring CT device spatial resolution based on the Catphan500 phantom.

Methods Prospectively collected quality control images from CT devices of different models across various hospitals, trained them using the YOLOv8 deep learning model, and compared the results with manual measurement methods to validate reliability.

Results The YOLOv8 deep learning model achieved a recall of 98%, a precision of 96%, and an mAP@0.5 of 0.988 on the test set, demonstrating excellent performance. The automated detection method measured a single image in an average of only 0.128 seconds, significantly outperforming manual measurement, with only slight differences observed at the extreme resolution.

Conclusion The YOLOv8 deep learning model provides a highly accurate and stable method for automated detection of CT spatial resolution based on the Catphan500 phantom, reducing manual intervention and measurement errors, and offering reliable technical support for the standardization and automation of CT device quality control.