SUMMARYIn this paper, a supervised edge detector based on a multilayer neural network, which is called a neural edge detector (NED), is proposed for detecting edges which coincide with edges traced by a human operator (e.g., a medical doctor). The NED is trained by use of the contours traced by a cardiologist. Using the trained NED, the contours coinciding well with the contours traced by a cardiologist are extracted from the left ventricular angiograms even with nonuniform contrast medium.The proposed contour extraction method consists of (1) detection of fine edges by the NED, (2) extraction of rough contours, and (3) contour tracing based on contour candidates synthesized from the rough contours and the edges detected by the NED. The contour of the left ventricle is automatically extracted by inputting two points manually. Experiments with clinical images show that the proposed method can stably extract the contours coinciding well with the contours traced by a cardiologist.