Pergamon zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA C o m p ute rie d Me d ic a l lm a g m g and Graphics. Vol. 19, No. I, pp. 47-99, 1995 C o p yrig ht 0 1995 Else vie r Sc ie nc e Ltd Printe d in the USA. A ll rig hts re se rve d 089.5-61 I l/Y5 $9.50 + 00 08956111(94)00036-O SUPERVISED INTERPRETATION OF ECHOCARDIOGRAMS WITH A PSYCHOLOGICAL MODEL OF EXPERT SUPERVISION’ Shriram Revankar,“’ David Sher,*’ Chris Cheung,* Valerie L. Shalin,+ Maya Ramamurthy,+ and Steve Rosenthal” *Department of Computer Science, State University of New York at Buffalo, 226 Bell Hall, Buffalo, NY 14260 ‘Department of Industrial Engineering, State University of New York at Buffalo, 342 Bell Hall, Buffalo, NY 14260 iEchocardiography-Graphics Laboratory, University of Alabama Hospital, Birmingham, AL 35213 (Received 9 May 1994) Abstract-We have developed a collaborative scheme that facilitates active human supervision of the binary segmentation of an echocardiogram. The scheme complements the reliability of a human expert with the precision of segmentation algorithms. In the developed system, an expert user compares the computer generated segmentation with the original image in a user friendly graphics environment, and interactively indicates the incorrectly classified regions either by pointing or by circling. The precise boundaries of the indicated regions are computed by studying original image properties at that region, and a human visual attention distribution map obtained from the published psychological and psychophysical research. We use the developed system to extract contours of heart chambers from a sequence of two dimensional echocardiograms. We are currently extending this method to incorporate a richer set of inputs from the human supervisor, to facilitate multi-classification of image regions depending on their functionality. We are integrating into our system the knowledge related constraints that cardiologists use, to improve the capabilities of our existing system. This extension involves developing a psychological model of expert reasoning, functional and relational models of typical views in echocardiograms, and corresponding interface modifications to map the suggested actions to image processing algorithms. Key Words: Echocardiography, Image semantics, Interactive image analysis. Medical image analysis, Perceptual and cognitive organization, Ultrasound, Visual attention INTRODUCTION Echocardiography is a popular clinical method for the identification and assessment of an entire spectrum of heart abnormalities (I). In recent years, visualization and quantitative analysis of the heart from two dimen- sional (2D) echocardiograms has received increased attention. It is possible to detect, quantize and visualize a large spectrum of heart abnormalities through seg- mentation of various regions in a two dimensional echocardiogram (2). However, automatic segmentation systems developed for quantitative analysis of echocar- diograms have not been successful in a clinical envi- ronment. Furthermore, owing to the complexity and importance of the task, human supervision of computer generated segmentation is often essential. ’ Correspondence should be addressed to Shriram Revankar, Webster Research Center, Xerox Corporation, Webster, NY, 14580. ’ Currently affiliated with Nassau Community College, Garden City, NY, I 1530. ’ Currently in private practice with Cardiology Associates of Atlanta, I I51 Cleveland Avenue, Suite D, East Point, GA, 30344. Two dimensional echocardiograms are images of cross sections of the heart, obtained by methodic regis- tering of the echoes generated by sound beam scans. The intensity of sound echoes are dependent on the type of the tissue, and change in the tissue type, at the point of backscatter. Echocardiograms record high intensity echoes at the points that correspond to the heart walls, valves and other tissue changes that lie within the scanning range. Therefore, it is possible to separate the wall regions from the blood volume by simple binary segmentation. Several researchers have used binary segmentation to extract the wall regions (2). However, because of the nature of imaging and complexity of the data, automatic methods may mis- classify some regions. To alleviate this drawback, re- searchers have proposed semiautomatic schemes that enable a user to interactively correct the misclassifica- tions (3, 4). In this paper, we study this problem of interactive refinement from the point of view of both image processing and human computer interaction, and develop a collaborative method for accurate segmenta- tion. 41