Pergamon Ultrasoundin Med.& Biol.,Vol.20, No. 1, pp. 1-10, 1994 Copyright © 1994 Elsevier Science Ltd Printedin the USA.All rightsreserved 0301-5629/94 $6.00 + .00 OOriginal Contribution ANALYSIS OF ULTRASONOGRAPHIC PROSTATE IMAGES FOR THE DETECTION OF PROSTATIC CARCINOMA: THE AUTOMATED UROLOGIC DIAGNOSTIC EXPERT SYSTEM A. L. HUYNEN, R. J. B. GIESEN, J. J. M. C. H. DE LA ROSETTE, R. G. AARNINK, F. M. J. DEBRUYNE and H. WIJKSTRA Department of Urology, University Hospital Nijmegen, P.O. Box 9101, 6500 HB Nijmegen, The Netherlands (Received 6 April 1993; in final form 2 August 1993) AbstractmThis paper describes a study on the automated analysis of ultrasonographic prostate images. With image processing, tissue characterization in the prostate was performed to assess the probability of malignancy. During prostate examinations, images were recorded at the positions where biopsies were taken. The used samples were divided into three groups. Two of them were used for the construction of a classification tree, and the third was used for the evaluation of this classification. A sensitivity of 80.6% and specificity of 77.1% were reached retrospectively. In a prospective way, these results were 80.0% and 88.2%, respectively. The prospective predictive value for cancer detection was 85.7%. The presented prospective value for image analysis was almost twice as high as the values normally found for prostate examination. Key Words: Ultrasound, Texture, Tissue discrimination, Prostate, Cancer detection. INTRODUCTION AND LITERATURE In the United States, prostate cancer is the second most diagnosed malignancy in men over 50 years old (Lee et al. 1989a; Waterhouse and Resnick 1989). It is also the most frequent male urological cancer with a grow- ing incidence, depending on age. When it is diagnosed in an early stage, however, prostatic carcinoma is cur- able (Lee et al. 1989a, 1989c; Shinohara et al. 1989). For the diagnosis of prostatism and the detection and staging of prostate cancer, ultrasound is employed by almost every urologist (Hodge et al. 1989; Lee et al. 1989a, 1989b; Scardino et al. 1989; Waterhouse and Resnick 1989; Zielke et al. 1985) because it is inter- active, easy to use, and a relatively cheap medical imaging technique. The quality of TransRectal Ultra- Sound (TRUS) has been improved markedly in the past few years, and an experienced urologist can detect suspicious lesions in the prostate with reasonable accu- racy (Schuster et al. 1986, 1987; Shinohara et al. 1989). However, some disadvantages are still attached to the use of TRUS (Bertermann et al. 1989; Chodak et al. 1986; Dahnert et al. 1986; Hendrikx et al. 1990; Lee et al. 1989c; Loch et al. 1990; Scardino et al. 1989; Address correspondence to: A. L. Huynen. Schuster et al. 1986; Shinohara et al. 1989; Waag et al. 1991; Waterhouse and Resnick 1989; Zielke et al. 1985). The appearance of cancer lesions in an ultra- sound image can vary, depending on the lesion type, location and transducer. Malignant tumours can appear anechoic, hypoechoic, or even isoechoic, and therefore not all cancers can be detected with TRUS. In early studies with 3.5 MHz transducers cancer was reported to be hyperechoic. According to Shinohara et al. (1989) this could be explained by the advanced stage of the disease that was found in these studies. In this ad- vanced stage of the disease, corpora amylacea could be responsible for the increased echodensity. Lee et al. (1989b) agreed with this explanation but also reported dystrophic calcification in necrotic tumour tissue to be responsible for the hyperechoic echopattern. With the current 7 MHz transducers with better resolution, ma- lignant lesions can be detected in an earlier stage, in which they appear anechoic, hypoechoic or isoechoic. D~ihnert et al. (1986) reported that a focal hypoechoic area, corresponding to a tumour, surrounded by a highly echogenic rim, represented a fibrous reaction in some of their cases. This could explain the positive findings of cancer in biopsies from hyperechoic regions alongside a hypoechoic tumour.