Full metadata record

DC Field Value Language
dc.contributor.authorSOJEONG CHEON-
dc.contributor.authorLee, Deukhee-
dc.date.accessioned2024-01-12T04:09:46Z-
dc.date.available2024-01-12T04:09:46Z-
dc.date.created2021-09-29-
dc.date.issued2020-11-
dc.identifier.issn--
dc.identifier.urihttps://pubs.kist.re.kr/handle/201004/77839-
dc.description.abstractFor the diagnosis and follow-up of scoliosis, back shape analysis using the anatomical landmarks is primarily used. However, finding the location of those landmarks is timeconsuming and clinician-dependent. This paper proposes a novel method for the automatic detection of anatomical landmarks on the 3D back surface scan. We use a public 3D body scan data set with many different postures, and an expert manually annotates the scan’s landmarks. Those landmarks consist of ten points and correspond to the structure of the spine. To pre-process the data set, we make a depth map from the scan with many different camera angles. Then, we crop the image to leave only the back surface area for our purpose and project the landmarks into the depth map. We train Stacked Hourglass Network that shows good performance in pose estimation and use the depth map and projected landmarks as input. After the landmark prediction, 2D points are reconstructed into the 3D space and we interpolate some of them to extract the spinal curve. We evaluate the trained network using the test data set and the computed prediction error shows that the proposed method can detect the anatomical landmarks successfully.-
dc.languageEnglish-
dc.publisherACCAS-
dc.subjectdeep learning-
dc.subjectautomatic anatomical landmark detection-
dc.titleAutomatic Detection of Anatomical Landmarks on 3D Back Surface Scan-
dc.typeConference-
dc.description.journalClass1-
dc.identifier.bibliographicCitationACCAS 2020-
dc.citation.titleACCAS 2020-
dc.citation.conferencePlaceJA-
dc.citation.conferencePlaceOn line-
dc.citation.conferenceDate2020-11-27-
dc.relation.isPartOfACCAS 2020-
Appears in Collections:
KIST Conference Paper > 2020
Files in This Item:
There are no files associated with this item.
Export
RIS (EndNote)
XLS (Excel)
XML

qrcode

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.

BROWSE