This paper introduces an automated method for recognizing and labeling leads in ECG images to create a machine learning dataset. We tested the main approaches on YOLOv11. For YOLOv11, we proposed two strategies: (1) detecting the entire bounding box containing all 12 leads in the image, and (2) detecting each lead symbol individually and determining the corresponding bounding box. The experimental results demonstrate that the YOLOv11-based method for detecting all 12 leads as a whole is the most effective, thanks to its fast and accurate detection capability. Along with preprocessing and data encoding techniques, this approach enables automated labeling, saving significant time and cost compared to manual methods. Experiments show that YOLOv11 achieves high performance in detecting and classifying ECG leads. This method not only optimizes the machine learning dataset creation process but also supports the development of intelligent healthcare systems and large-scale cardiovascular research.