A combined local-global optical flow approach for cranial ultrasonogram image sequence analysis

Tissue-motions can represent the strength of artery pulsation of a newborn baby for pediatrics diagnosis. In this paper, we propose a novel method to estimate the tissue-motion quantitatively by combining local and global motion estimation methods, namely combined local-global (CLG) optical flow technique in cranial ultrasonogram of newborn babies. Tissue motions are estimated in different coronal sections successfully. Further, we analyze the time variant tissue-motion by using discrete Fourier transform. We also observed the pulsation in the time variant motion images and strong pulsation is observed in the harmonic frequency of tissue-motion that has a relation to the heartbeat frequency of a newborn baby.

I. Introduction

Ultrasonogram is routinely used at the various sites of medical diagnosis because of its convenience, easy to use, low cost in compared with magnetic resonance imaging (MRI) and X-ray computed tomography (CT). Cranial ultrasound scans are essential part of the routine investigation during neonatal intensive care. The scan ultrasound image sequences are not only use for real-time diagnosis but also recorded for future diagnosis. Pediatricians use an ultrasonogram to find the cranial abnormalities in a newborn baby, since they can observe the brain tissue through the anterior fontanel [2], [5]. A key point of diagnosis is whether there is the artery pulsation accompanied with blood flow or not [5], [6]. However, there are a few attempts were taken to analyze the tissue-motion from the moving images of ultrasonogram for pediatrics diagnosis. In order to detect artery pulsation from a series of noisy ultrasound echo images of newborn baby head for pediatric diagnosis, a digital image processing system was developed by using the algorithm based on Fourier transform [6]. In order to observe artery pulsation in neonatal cranium at the site of diagnosis, a real time processing system was developed for continuous display and detection of artery pulsation from moving images of neonatal cranial ultrasonogram [10]. The entire algorithms used above were mainly based on the time sequence variation of pixel value (intensity) in a series of ultrasound echo images. However, the pixel value can easily vary. Therefore, the above methods were very sensitive to the brightness of the original ultrasonogram movie. In this paper, we calculate the tissue-motion due to artery pulsation in cranial ultrasound image sequences by using CLG optical flow technique, which is not sensitive to image brightness change. This is very helpful for quantitative characterization of artery pulsation. Moreover, the time variant tissue-motions are analyzed by using Fourier transform, emphasizing that the tissue-motion at a particular frequency has a relation with heartbeat frequency of newborn baby.