Edge detection in medical images with quasi high-pass filter based on local statistics

Abstract

Abstract We developed a robust, quasi high-pass filter for edge detection in medical images. The kernel-based algorithm of our detector was similar to that of conventional edge detectors. The proposed edge detector has a mathematical form of local variance and is adaptive in nature. The mathematical formulation of the detector was exploited and re-expressed as a quadratic form of the Toeplitz matrix. The detector has a highly structured internal architecture with abundant spatial isotropic symmetricity. The proposed operator can greatly reduce problems frequently encountered in edge detection including fragmentation, position dislocation, and thinness loss. The detector is robust to noise and can efficiently extract crucial edge features. We named this new operator as the WL operator (Wang and Lin). The performance of the WL operator was compared to that of other edge detectors by using Pratt’s figure of merits. In addition, the performance was confirmed with experts by using visual analog scale scores. The results obtained using the WL operator for different medical imaging modalities including X-ray, CT, and MRI are promising. Therefore, the WL operator warrants further investigation.