Anautomated pixel classification methodusing surface expansion Application toMRI imagesequence
Antonio Pinti, Patrick Hédoux, Kang Han, Abdelmalik Taleb‐Ahmed · 2006
Thispaperdescribes an automatedpixel classification methodusingsurfaceexpansion. The originality ofthis workresides inthedefinition anduseof smallpictures (called imagelet) ofincreasing size centered onthepixel ofinterest. Thisallows fortheextraction ofaset oflocalandglobal parameters associated tothepixel investigated. Thissetofparameters thenpermits bodytissue separation. Classification wasobtained using a multilayer artificial neural network. Thenewapproach proposed wasapplied tomainlowerlimb tissue classification inMRI imagesequences. Fourkindsof bodytissue weretakenintoconsideration inthisstudy (muscle, adipose tissue, cortical boneandspongybone). A database consisting of1400prototypes wascreated inorder toevaluate thismethod's performances. Theclassification success ratewasfoundtobe87%.Thismethodtherefore provedtobereliable androbust toanalyze MRI image sequences in20 lowerlimbs, representing about2000 pictures.