1 Using Phase Information in Ultrasound RF-Signals for Tissue Characterization Ivana Despotovi´ c, Bart Goossens, Ewout Vansteenkiste, Aleksandra Pižurica, Wilfried Philips Ghent University, Department of Telecommunication and Information Processing (TELIN) - IPI - IBBT E-mail: ivana.despotovic@telin.ugent.be Abstract—Due to many processing steps, the B-mode ultra- sound images commonly used for medical diagnosis, contain less information than the original Radio-Frequency (RF) signals received by the ultrasound probe. Moreover, the raw RF-signals contain potentially more information on the investigated tissue. In this work we explore whether phase information of the ultrasound RF-signals can contribute to a better characterization of insonified preterm brain tissue. We use time series of raw RF-signals obtained with a linear probe. Amplitude and phase information are extracted from complex valued demodulated RF- signals and subsequently an envelope image is reconstructed. Then, we analyze the distribution of the phase differences both within the signal and between neighboring signal scan lines and compute the entropy of those phase differences in local neigh- borhoods. Our initial qualitative results indicate that statistical properties of the phase differences, such as the proposed entropy, bring useful information about the insonified tissue. Index Terms—Ultrasound, Radio-Frequency signals, Phase information I. I NTRODUCTION The Radio-Frequency (RF) signal is the unprocessed elec- trical signal coming from the ultrasound scanner’s probe. It thus contains all the information on the propagation of the acoustic waves and their intersections with the scanned tissue. The B-mode ultrasound image, displayed on the scanner screen, is a brightness gray-scale image that represents the envelope of the RF signals. Such an image is obtained by demodulating the RF signals. Note that before the image is actually displayed, several processing steps are applied like time-gain compensation, filtering, rectification and log- compression. Because these steps differ on each ultrasound device, images taken with different machines are difficult to compare. Due to these processing, the loss of information disables us to take full advantage of the received signal in terms of signal processing and automatic diagnoses. In recent years there have been many studies in the field of ultrasound tissue characterization using unprocessed RF signals, which are recognized as a rich source of information [1]-[5], [8], [11]. However, the main difficulty in ultrasound RF-signal processing is the fact that the interaction between tissue and sound waves is poorly understood. Little is known on how changes in tissue characteristics influence the reflected signal. Most of the existing works in this area address statis- tical properties of the RF-signals and envelope images [1]- [4]. Also, little analysis of the phase properties in ultrasound images has appeared, mostly because of the random nature of the phase information. In this paper we show that, even though the phase infor- mation is mostly randomly distributed, there exist correlations between neighboring RF-signal samples that can be used for tissue characterization. We propose the entropy of the phase difference as a statistical measure for the phase correlation and investigate its relation to the amplitude information. The paper is organized as follows: Section II presents materials and methods that are used in this paper. Firstly, the experimental set-up of the RF-signal acquisition is discussed. Secondly, the demodulation of the RF signals and the image reconstruction process is explained. Next, we briefly describe the entropy calculation of the phase difference. In section III results are displayed together with the visualization of the fused envelope (amplitude) and entropy image using the HSV color space and a discussion is given. Finally, the conclusion is presented in section IV. II. MATERIALS AND METHODS A. Experimental data For the ultrasound data collection we use a Picus ultrasound machine (ESAOTE NV, Maastricht) that contains an integrated PC and the capability of collecting and recording raw RF- signals. These signals are acquired with the Art.Lab software equipped with a data acquisition card, where each acquisition sample consists of 16 bits with the upper 12 bits used for RF-samples and lower 4 bits used for event coding, e.g. trigger signals. We use a linear ultrasound probe (type L10-5 40mm) which supports 5/7.5/10 MHz frequencies. Acquired RF-signals are sampled according to their frequency content, i.e. three to four times the dominant carrier frequency and afterwards stored in real-time in the internal memory of a personal computer system as an RF-matrix. The time interval of the RF-matrix depends on the Pulse Repetition Frequency (PRF) of the probe (△t=1/PRF), while the depth interval 314