Automatic corner detection of the cervical vertebrae C2-C7 in fluoroscopic recordings
Author
Jakobsen, Ida Marie Groth
Term
4. term
Publication year
2019
Submitted on
2019-01-03
Pages
45
Abstract
Nakkesmerter har store økonomiske konsekvenser. En måde at finde årsagen på er at undersøge, hvordan nabohvirvler i nakken roterer og glider i forhold til hinanden. Røntgenvideo (fluoroskopi) giver mange løbende billeder og kan derfor afspejle denne bevægelse mere præcist end enkelte stillbilleder. For at måle bevægelsen skal man normalt finde fire anatomiske landemærker på hver hvirvel: hjørnepunkterne på hvirvellegemet. Disse markeringer laves ofte manuelt, hvilket er tidskrævende. I dette studie var målet at finde landemærker på halsryghvirvlerne C2–C7 automatisk i fluoroskopiske optagelser. Fire forsøgspersoner bøjede og strakte nakken (fleksion og ekstension), mens de bar briller med eksterne markører, som angav baghovedets (occiputs) position. Algoritmen lokaliserede centerpunkterne for markørerne og de fire hjørner på C3–C6 samt de nedre hjørner på C2 og de øvre hjørner på C7. Metoden byggede på multisegmentering, en billedanalysestrategi der opdeler billedet i områder, og blev testet på fem enkeltbilleder (frames) fra hver fluoroskopisk optagelse. De eksterne markører blev fundet korrekt i 27 af 40 testede frames, og hjørnerne på halsryghvirvlerne blev fundet i 31 frames. Resultatet viser, at cervikale hvirvler og centrale landemærker kan opdages uden manuel indblanding.
Neck pain has a large economic impact. One way to identify its causes is to examine how neighboring neck bones rotate and slide relative to each other. X-ray video (fluoroscopy) provides many consecutive images and can capture this motion more accurately than single snapshots. Measuring motion typically requires locating four anatomical landmarks on each vertebra: the corner points of the vertebral body. These are often marked by hand, which is time-consuming. This study aimed to automatically detect landmarks on cervical vertebrae C2–C7 in fluoroscopic recordings. Four participants performed neck flexion and extension while wearing glasses with external markers to indicate the position of the occiput (back of the head). The algorithm identified the centers of the external markers and the four corners of vertebrae C3–C6, as well as the lower corners of C2 and the upper corners of C7. The method used a multi-segmentation approach, a computer vision technique that separates the image into regions, and was tested on five frames from each fluoroscopic recording. The external markers were correctly detected in 27 of the 40 tested frames, and the vertebral corners were detected in 31 frames. These results show that cervical vertebrae and key landmarks can be detected without manual input.
[This abstract was generated with the help of AI]
Documents