Automatic identification of landmarks for standard slice positioning in brain MRI

Abstract

To demonstrate a novel automatic slice-positioning technique based on three new anatomical landmarks and to standardize prospective scans by lowering rotational and translational variances. After defining the interpeduncular fossa corner and the eyeball centers as landmarks, they are manually labeled on 25 different T1 MRI scans. New scans are produced according to the Eyeball centers-Mesencephalon (EM) plane. The comparison of angular deviations at EM and original scans is based on the comparison of rotational angles according to manually labeled Talairach points on both scans. The same variability comparison is also done with automatically captured landmarks to see the effects of segmentation errors. Analysis of variances proved significant lowering of intersubject variability for pitch and yaw angles (P(pitch) < 0.005, P(yaw) < 0.001), which are the two basic causes of misalignments. Automatic segmentation accuracy is proved by paired t-test and significance tests. A new field of view and slice orientation proposed by the EM technique will have fixed the follow-up scans by significantly lowering the rotational and translational variances. The EM technique will precisely match the intrasubject scans and produce better standardized intersubject scans. The distinguishing features of landmarks are sufficient for robust automatic capture.