Ultrasonic Tissue Characterization – Assessment of Prostate Tissue Malignancy in vivo using a conventional Classifier based Tissue Classification Approach and Elastographic Imaging A. Lorenz 1 , A. Pesavento 1 , U. Scheipers 2 , H. Ermert 2 , M. Garcia-Schürmann 3 , H.-J. Sommerfeld 3 , T. Senge 3 , S. Philippou 4 1 Lorenz & Pesavento IT, Bochum, Germany, www.lp-it.de 2 Dept. of Electrical Engineering, Ruhr University, Bochum, Germany 3 Dept. of Urology, Ruhr University, Bochum, Germany 4 Dept. of Pathology, Ruhr University, Bochum, Germany Abstract - In this paper we present the development of a combined system which is able to exploit the benefits of two methods used for tissue characterization, strain imaging and tissue classi- fication using a trainable classification system. Our system is able to acquire in vivo multi-compression rf-data for the calculation of the tissue strain, i.e. the elastic properties of tissue, induced by tissue compression. At the same time a Neuro-Fuzzy classification system is used to map the tissue malignancy. In vivo Classification results and in vivo strain images are presented. The images of the two new modalities are compared to demonstrate the advantages and restrictions of both methods. INTRODUCTION In the past, several ultrasound methods have been described to assess the malignancy of prostate tissue. Two major approaches are the investigation of tissue elasticity by strain imaging [1] and the detection of malignantant tissue areas by tissue classification with a trainable classification system [2]. Ultrasonic strain imaging refers to the visualization of tissue elasticity for medical diagnosis. With this technique small displacements between ultrasonic image pairs which are acquired under varying axial compression are determined using a cross- correlation analysis of corresponding a-lines of an rf-data set. The derivative of the displacement field is equal to the strain in the tissue. Tumors often can be detected by palpation, therefore strain imaging promises to yield good results to detect such tumors. For multicompression strain imaging a sequence of rf-images is acquired under step-wise increasing compression in order to extend the dynamic range and the resolution of the strain estimates [3]. Due to the lateral motion of the insonified object with respect to the axial beam direction the use of a sector probe leads to significant motion artifacts even in a plane strain state. A fast and efficient method for the correction of lateral motion artifacts is described in [4,5]. An efficient method for the fast calculation of strain images is described in [6]. The methods [5] and [6] were used in this paper. Tissue classification means the segmentation of image data into small segments and the calculation of statistical tissue parameters either obtained from spectrum analysis or texture analysis of the ultrasonic echo data. The parameters are used in combination with known histological findings to construct a classification system which is able to determine the malignancy state within a region of interest [7,8].