Detecting visual differences in reconstructed images using a region-based test for outliers

The objective of this work is to present a region-based image change detection algorithm based on a statistical test for outliers. The performance of the proposed algorithm is compared to the performance of another algorithm we have developed, which is based on the Grubbs statistical test for outliers. The two algorithms are tested on pairs of reconstructed images obtained from realistic Monte Carlo simulations of a real patient PET scan. Simulation data is reconstructed with the Software for Tomographic Image Reconstruction using two image reconstruction algorithms Filtered Back-Projection 3D Reprojection (FBP) and Ordered Subset Maximum a Posteriori One Step Late Algorithm (OSMAPOSL). The results of the two algorithms for the cases in which the compared images do and do not contain visual differences are presented. Images reconstructed with one and the same algorithm, as well as with different algorithms are compared. It is found that the region-based approach detects more false alarms when the compared images are reconstructed with FBP, but outperforms the standard approach when the images are reconstructed with OSMAPOSL. The region-based approach is found to be more sensitive to high gradient regions in the difference images.