Skip to main content
. 2022 Feb 17;22(4):1547. doi: 10.3390/s22041547

Figure 4.

Figure 4

Proposed Framework: Dataset is augmented, annotated, and passed through the YOLOv5 to obtain localized vertebrae. L1 and S for LLA and L5 and S for LSA are extracted from the localized vertebrae, which are then passed through the HED U-Net to obtain the edge-based segmentation. Images are smoothed using Gaussian smoothing filter, and Harris corner detector is applied to obtain the corners of desired vertebrae. LLA and LSA can be found from the corners.