Semi-automatic method for delineation of midbrain in transcranial ultrasound images

Abstract

The aim of this paper is to present new method developed for semi-automated segmentation of midbrain in low quality transcranial ultrasound images combining principle of statistical shape modelling and local phase accumulation based boundary detection strategy. Averaged results of 142 cases obtained comparing contours extracted by proposed method and annotations: Hausdorff distance - 3.36±0.88 mm, mean squared error - 2.18±0.55 mm, Dice coefficient - 0.88±0.03. An efficient technique for midbrain segmentation was proposed. The combination of experience based shape model and intensity-amplitude invariant edge detector was applied for extraction of fuzzy boundaries of midbrain in TCS image.