Fetal skull analysis in ultrasound images based on iterative randomized Hough transform

Abstract

Analysis of ultrasound fetal head images is a daily routine for medical professionals in obstetrics. The contours of fetal skulls often appear discontinuous and irregular in clinical ultrasound images, making it difficult to measure the fetal head size automatically. In addition, the presence of heavy noise in ultrasound images is another challenge for computer aided automatic fetal head detection. In this paper, we first utilize the stick method to suppress the noise and compute an adaptive threshold for fetal skull segmentation. Morphological thinning is then performed to obtain a skeleton image, which is used as an input to the Hough transform. Finally, automatic fetal skull detection is realized by Iterative Randomized Hough Transform (IRHT). The elliptic eccentricity is used in the IRHT to reduce the number of invalid accumulations in the parameter space, improving the detection accuracy. Furthermore, the target region is adaptively adjusted in the IRHT. To evaluate the performance of IRHT, we also developed a simulation user interface for comparing results produced by the conventional randomized Hough transform (RHT) and the IRHT. Experimental results showed that the proposed method is effective for automatic fetal head detection in ultrasound images.