 |
Now, let's look at the details of our "Data Preprocessing" phase.
The first step involved extracting and carefully analyzing DICOM files to identify the slices that corresponded to the relevant vertebral regions. Next, based on the segmentation masks, bounding boxes were skillfully generated to encapsulate the identified bones within each slice.
To seamlessly integrate this information into the YOLOv5 framework, the bounding box coordinates were extracted and thoughtfully saved in a TXT file, ensuring compatibility with the YOLOv5 model. Additionally, for enhanced efficiency in model feeding, the corresponding DICOM images underwent conversion to the PNG format.
This comprehensive data preprocessing pipeline lays the foundation for accurate and effective utilization of our dataset in training the YOLOv5 model for cervical spine bone detection.
|
IEEE/ICACT20240327 Slide.08
[Big Slide]