Automated design of optimal border detection criteria: learning from image segmentation examples

Manual analysis of ever increasing numbers of diagnostic medical images is tedious and impractical in a clinical setting. Employment of automated image segmentation approaches is increasingly common. Unfortunately, the utility of existing medical image analysis systems is limited by their narrow, highly specific task orientation. We have developed a method for an automated design of optimal border detection criteria based on learning from image segmentation examples. Two learning approaches were proposed: A feature-based method using direct least square error minimization and a radial basis neural network. The two approaches were validated in simulated ultrasound images, and in intracardiac and intravascular ultrasound images. The achieved performance was comparable to that of our previously reported single-purpose border detection methods. Our approach facilitates development of general multipurpose medical image segmentation systems that can be trained for different types of image segmentation applications. Such systems would considerably simplify the task of border detection in the rapidly changing world of medical imaging.