IEEE TRANSACTIONS ON ULTRASONICS, FERROELECTRICS, AND FREQUENCY CONTROL, . 58, . 11, NOVEMBER 2011 2344 0885–3010/$25.00 © 2011 IEEE A Restoration Framework for Ultrasonic Tissue Characterization Martino Alessandrini, Simona Maggio, Jonathan Porée, Luca De Marchi, Nicolo Speciale, Emilie Franceschini, Olivier Bernard, and Olivier Basset Abstract—Ultrasonic tissue characterization has become an area of intensive research. This procedure generally relies on the analysis of the unprocessed echo signal. Because the ultra- sound echo is degraded by the non-ideal system point spread function, a deconvolution step could be employed to provide an estimate of the tissue response that could then be exploited for a more accurate characterization. In medical ultrasound, de- convolution is commonly used to increase diagnostic reliability of ultrasound images by improving their contrast and resolu- tion. Most successful algorithms address deconvolution in a maximum a posteriori estimation framework; this typically leads to the solution of ℓ 2 -norm or ℓ 1 - norm constrained opti- mization problems, depending on the choice of the prior distri- bution. Although these techniques are sufficient to obtain rel- evant image visual quality improvements, the obtained reflectivity estimates are, however, not appropriate for classifi- cation purposes. In this context, we introduce in this paper a maximum a posteriori deconvolution framework expressly de- rived to improve tissue characterization. The algorithm over- comes limitations associated with standard techniques by using a nonstandard prior model for the tissue response. We present an evaluation of the algorithm performance using both com- puter simulations and tissue-mimicking phantoms. These stud- ies reveal increased accuracy in the characterization of media with different properties. A comparison with state-of-the-art Wiener and ℓ 1 -norm deconvolution techniques attests to the superiority of the proposed algorithm. I. I M  ultrasound is widely employed in the clinical routine to assess possible abnormalities in several parts of the human body. Currently, the diagnosis relies almost exclusively on the visual observation of the ultra- sound sequences, but it has been widely reported that computer analysis of the echo signal can be employed to infer diagnostically relevant information on the tissue state which is otherwise imperceptible from simple visual inspection. This observation motivated the development of computer-aided detection (CAD) tools to function as a support to the physician in the interpretation of ultra- sound scans and to guide the physician in the decision- making process in the case of suspicious situations. The use of CAD tools on clinical data has led to relevant re- sults in several applications, such as prostate cancer de- tection on trans-rectal ultrasound images [1], detection of suspicious masses in breast ultrasound [2], and diagnosis of hepatic steatosis [3]. The output of CAD systems is derived from the quan- titative analysis of the echo signal. In this context, a large number of features of different natures have been proposed in literature; these can be subdivided according to their contribution in highlighting specific tissue properties. Tis- sue characterization based on the acoustic parameters such as attenuation and backscattering coefficients ex- tracted from RF echo signals has been widely studied. These quantities are commonly estimated by using 1-D [4] or 2-D [5] spectral analysis of the RF signal. Spectral features have proven to provide useful output for diagnosis of diseases in various organs, such as the eye, prostate, breast, and liver; see [6] for a comprehensive review. In addition to RF-spectrum analysis, many researchers have used texture features extracted from ultrasound B-scan images for characterization purposes, because the speckle pattern in the ultrasonic image can reveal structural in- formation about the tissue. Usefulness of textural features within clinical settings has been widely documented, e.g., for prostate carcinoma diagnosis [7], [8], evaluation of liver diseases such as hepatoma and cirrhosis [9], and detection of atherosclerotic plaques in the carotid artery [10]. A third class of features derives from modeling the echo signal amplitude distribution by means of suitable para- metric probability density functions (pdfs). These statisti- cal features have been shown to be well-related to scatterer concentration and distribution pattern. A variety of mod- els have been proposed in literature. The most popular model is represented by the Rayleigh distribution for the envelope signal, which can be analytically derived for dif- fusive scattering (or fully developed speckle) regions [11, pp. 48–50], [12]. Nevertheless, diffuse scattering conditions are often violated in biological tissues, either because the number of scatterers per resolution cell may not be large enough, or because of the presence of regular patterns in the scatterers location. In these cases, more complex models must be adopted, such as Rician [11, pp. 50–52], K [13], Homodyne-K [14], or Nakagami distribution [15]. In Manuscript received October 22, 2010; accepted August 25, 2011. M. Alessandrini is with the Advanced Research Center on Elec- tronic Systems for Information and Communication Technologies E. De Castro (ARCES), Università di Bologna, Bologna, Italy (e-mail: martino.alessandrini@creatis.insa-lyon.fr). S. Maggio, L. De Marchi, and N. Speciale are with the Dipartimento di Elettronica, Informatica e Sistemistica (DEIS), Università di Bologna, Bologna, Italy. J. Porée, O. Bernard, and O. Basset are with Centre de Recherche et d’Applications en Traitement de l’Image et du Signal (CREATIS), Cen- tre National de la Recherche Scientifique (CNRS) UMR 5220, Institut National de la Santé et de la Recherche Médicale (INSERM) U630, Uni- versité de Lyon, Institut National des Sciences Appliquées (INSA)-Lyon, Villeurbanne, France. E. Franceschini is with the Laboratoire de Mécanique et d’Acoustique (LMA), Centre National de la Recherche Scientifique (CNRS), UPR 7051, Marseille, France. Digital Object Identifier 10.1109/TUFFC.2011.2092