Vertebrae Detection Algorithm in CT Scout Images

Document Type

Conference Proceeding


In order to solve the tedious and time-consuming works for CT scan planning manually, we proposed an automatic detection method of vertebrae in CT scout images. In this method, firstly, HOG features of the training samples were computed, which were imported into the random forest classifier for training. Then we rotated the CT scout images seven times for detecting multi-angle vertebrae. The trained classifier was employed to detect the vertebrae in test images. Finally, we merged the detection results with overlapping regions. For 76 images, experimental results show that the sensitivity of vertebrae detection by our method reached 95.18 % with 0.96 false positive per image.

UNLV article access