Fetal Brain Segmentation using Improved Maximum Entropy Threshold

Abstract

Fetal Magnetic Resonance Imaging (MRI) helps in learning about fetal brain development and has many advantages over ultrasound imaging technique. Such studies require segmentation of fetal brain. Manually segmenting the developing brain of a fetus is a challenging task and it necessitates anatomical knowledge. Hence, the proposed automatic method segment fetal brain region from fetal MRI. The pipeline of this proposed method comprises diffusion, morphological filtering, thresholding and connected component analysis. The proposed method is validated using sixteen retrospective T2-weighted fetal brain volumes. The segmented portions obtained by the method are compared with manually segmented gold standard, both qualitatively and quantitatively by estimating the Dice (D), Sensitivity (S), Specificity (Sp) and Hausdorff distance (HD). A highest value of 0.9268 for D, 0.9790 for S and 0.9983 for Sp are obtained by the method. For contour overlap, the method produced lowest value of 3.3 mm for HD value. Thus, our automatic algorithm to segment fetal brain from MRI gives competitive results compared to that of existing methods.