Markov Random Field Segmentation Based Sonographic Identification of Prenatal Ventricular Septal Defect.

Ventricular Septal Defect (VSD), a hole in chamber wall that separates right ventricle and left ventricle, is the most common congenital heart defect (CHD) among fetuses of India with 33% of occurrence range. In this paper, we present a method for image analysis to identify the VSD from the fetal heart 2 dimensional ultrasound images. Ultrasonography is the safest and essential imaging modality to infer the growth status of the fetus in womb. Generally making diagnosis from the fetal heart ultrasound images is a most difficult task for the physicians because of the image is highly corrupted with speckle noise and moreover the anatomical structure identification of fetal heart remains a big issue due to the fast pumping nature of fetal heart. The boundaries of ultrasound images appears with irregular edge structures due to the inconsistent appearance of speckle noise, and hence conventional segmentation approaches based on pixel intensity fails to delineate the clinical ultrasound structure boundaries. The sonographic marker used clinically to identify VSD is the H-shaped symbol seen in the ultrasound fetal heart 4 chamber view image plane. We combine a robust pre-processing methodology and segmentation approach based on unsupervised Markov Random Field (MRF) model to highlight the sonographic marker for VSD screening from the 2 dimensional ultrasound images. The experimental result shows that the average segmentation error obtained for this combined approach was around 9.83% while compared with the results of expert sonographologist. [ABSTRACT FROM AUTHOR]