Automatic segmentation on CT scans of humain brain
Author
Utasi, Tamás
Term
4. term
Publication year
2009
Pages
68
Abstract
This thesis investigates automatic segmentation of human brain CT scans into brain tissue, cerebrospinal fluid, skull, and background (including calcifications). Although MRI distinguishes soft tissues better, CT remains relevant due to MRI contraindications (metal implants), wider availability, and considerations such as claustrophobia. The work applies fuzzy c-means (FCM) clustering as an unsupervised method and explores extended feature vectors, including intensity histogram moments and texture features derived from co-occurrence matrices, as well as a Population-Diameter Independent (PDI) modification. Implementation, quality handling, and evaluation are described, with results compared against manual segmentations. Current findings indicate that FCM, even with the PDI extension, is not optimal for this task because it tends to overclassify cerebrospinal fluid as brain tissue. The thesis concludes with limitations and potential improvements.
Dette speciale undersøger automatisk segmentering af CT-scanninger af den menneskelige hjerne i hjernevæv, cerebrospinalvæske, kranium og baggrund (inklusive forkalkninger). Selvom MR bedre kan adskille blødt væv, er CT fortsat relevant på grund af kontraindikationer for MR (metalimplantater), bred tilgængelighed og hensyn til klaustrofobi. Arbejdet anvender fuzzy c-means (FCM) klyngealgoritmen som en usuperviseret metode og afprøver udvidede featurevektorer, herunder intensitets-histogrammomenter og teksturtræk fra sammenforekomstmatricer, samt en Population-Diameter Independent (PDI) udvidelse. Implementering, kvalitetshåndtering og evaluering beskrives, og resultaterne sammenlignes med manuelle segmenteringer. De aktuelle resultater viser, at FCM, selv med PDI, ikke er optimal til opgaven, da cerebrospinalvæske ofte overklassificeres som hjernevæv. Specialet afslutter med begrænsninger og mulige forbedringer.
[This apstract has been generated with the help of AI directly from the project full text]