Calculation of true T1, T2 and proton density images for the elimination of signal intensity artifacts in segmentation of brain tissue in magnetic resonance imaging

Abstract

Tibbi goruntulerde dokularin segmentasyonu radyologlara sagladigi yararlar baki-mindan onemli bir konudur. Radyologlar bu yontemi hastaliklari teshis etmede tumorleribelirlemede ve dejeneretif hastaliklarin takibinde kullanabilir. Manyetik RezonansGoruntuleme de segmentasyon sirasinda karsilasilan sorunlardan bir tanesi manyetikalandaki bozukluklardir. RF coil deki alan bozuklugu nedeniyle alinan sinyal yogunlu-gundaki degisim gorunrunun parlaklik degerlerini etkiler. Bu etki belli bir gorunrulemesistemi icin segmentasyonda problem yaratabilir. Birden cok goruntuleme sistemindekullanilacak bir algoritma dusunulecek olursa doku parlakliklarinda ve RF coil alanibozukluklarindaki degisim nedeniyle hata orani artabilir. Bu alismada ilgili problem-lerin cozulmesi icin hesaplanmis T1, T2 ve proton yogunlugu degerlerini kullanan biralgoritma onerilmektedir. Bu parametreler ayni bolgenin farkli dort parametre ilegoruntulenmesi ve Levenberg-Marquardt Methodu nun uygulanmasi ile hesaplanmistir.Daha sonra yapilan sinifandirma sonucunda segmentasyon sonucu elde edilen goruntulerolusturulmustur. Uc farkli MR cihazinda goruntulenen yedi farkli gonullunun beyingoruntuleri Gri madde, Beyaz Madde ve Serebrospinal sivi olarak ayristirilmistir. Dokuayristirma agirlikli T1, T2 ve PD goruntuler kullanilarak ve hesaplanms gercek T1, T2PD goruntuleri kullanilarak yapilmitir. Karlastirmalar kesitler aras, goruntuleme cihazlarıarası ve gonulluler arası yapılmıs ve sonuc olarak hesaplanmıs T1, T2 ve PD goruntuleri kullanıldıgındabeyin dokularnn ayrtrlmasnda olduka iyi iyiletirme salanmistir.

Segmentation of tissues in medical imaging is an essential subject because ithelps the radiologists to be able to identify diseases, tumors and follow the degenera-tive diseases. In Magnetic Resonance Imaging (MRI) one factor that causes a problemduring segmentation is the inhomogeneity in the magnetic ¯eld. Mainly the RF coilinhomogeneity e®ect causes intensity inhomogeneity through the image. This intensityinhomogeneity may cause segmentation algorithms to fail for a speci¯c imager system.In case an algorithm that can be used in many imagers is needed the di®erence in thetissue intensities and the RF coil inhomogeneity change may cause greater failures.To overcome this problem a method which uses calculated T1, T2 and proton densityparameters is proposed. These parameters are calculated from MRI images using foursampling points (four sets of images of the same region with di®erent parameters)and using Levenberg-Marquardt Method. Then maximum likelihood classi¯cation isapplied to distinguish the tissues and the segmented images were constructed. GrayMatter, White Matter and Cerebrospinal Fluid were segmented in MR brain imagesof seven volunteers. The subject heads were scanned with three di®erent MR im-agers. Tissue segmentation was performed with the weighted T1, T2 and Proton Den-sity images along with the computed true T1, T2 and PD. Comparisons across imageslices; across imagers and across subjects indicated that signi¯cant improvement canbe achieved if the computed T1, T2 and PD images are used for the segmentation ofbrain tissue.