基于关键点定位的脊柱冠状位Cobb角自动测量方法

打开文本图片集

中图分类号：TN919-34；TP391 文献标识码：A 文章编号：1004-373X（2025）15-0127-08

Keypoint positioning based automaticmeasurement of Cobb angle in spine coronal position

JIANGPeng1，TANGYu'，HEZhiqin1，WUQinmu1，WANGLihang2 Colegei;i

Abstract：Scoliosisisaseriousdiseasethatimpacts humanphysicalhealthsignificantlyandtheCobbanglecanreflectthe severityof scoliosis.Inviewofthetime-consumingmanual measurement，largeerrorandlowaccuracyofCobbangle measurement intheexisting deeplearning methods，acoronal Cobbangleautomaticmeasurementmethodbasedon keypoint positionigisproposed.Inthismethod，akeypointlocalizationnetworkbasedoncenterwithofsetisusedtolearntheoffset betwenthevertebralvertexandcenterpoint，enablingvertebralvertexcordinateregresion.Supervised trainingisalso conductedbydjacentvertebralofsetstoutilzevertebalpositioniformationfullandachievepreciselocalizatio.Finalypretraining isconductedontheCOCOdatasetbytransferlearning.Theresultsof experimentsonlocalandopendatasets demonstrate thattheproposedmethodcan locatevertebralcenterandvertexaccurately.Inadition，the meanabsoluteerrors （MAEs） between the measured Cobb angles of the three segments and the true values are 3.0∘ ， 2.5∘ ，and 2.4∘ ，respectively，and thePearsoncorelationcoeficientsisO.94，0.95，and0.95.Tosumup，theproposed methodcanrealize theautomatic measurement of Cobb angles accuratelyand reliably.

Keywords：scoliosis;Cobbangle;keypoint positioning;centerpointofset;ofsetof adjacent vertebrae;transferlearning

0 引言

近年来，为实现自动的Cobb角测量以及减小测量误差，采用深度学习的方法进行Cobb角自动测量成为了研究热点，其主要分为两类，即基于分割模型的方法和基于坐标回归的关键点检测方法。（剩余10078字）