Advertisement
Published in:
2020 | OriginalPaper | Chapter
Accurate Automated Keypoint Detections for Spinal Curvature Estimation
Authors : Kailin Chen, Cheng Peng, Yi Li, Dalong Cheng, Si Wei
Published in: Computational Methods and Clinical Applications for Spine Imaging
Publisher: Springer International Publishing
Activate our intelligent search to find suitable subject content or patents.
Select sections of text to find matching patents with Artificial Intelligence. powered by
Select sections of text to find additional relevant content using AI-assisted search. powered by
Abstract
In order to estimate the spinal curvature, we propose two methods to detect the spinal keypoints at first. In Method-1, we use a RetinaNet to predict the bounding box of each vertebra followed by a HR-Net to refine the keypoint detections. In Method-2, we implement a similar two-stage system, which firstly extract 68 rough points along the spine curves using a Simple Baseline. We then generate patches and make sure each of them contains three vertebrae at most based on ground truth. We train a second Simple Baseline to predict the exact keypoints of these patches, which are generally not fixed in numbers. A delicate postprocess of clustering is proposed to deal with dense keypoint predictions due to the freedom of patch selections. After fusing Method-1 and Method-2, we achieve competitive results on the public leaderboard of AASCE2019 challenge.