A novel outlier detection based approach to registering pre- and post-resection ultrasound brain tumor images
Haradhan Chel, Prabin Kumar Bora · 2017
Brain tumor surgery is performed with the help of both pre-operative and intra-operative images. Ultrasound(US) can effectively be used as an intra-operative modality. To avoid inconvenience caused by the surgical intervention and the brain shift, the pre-operative US image is needed to be registered to the intra-operative US image. This registration process does not perform well in the presence of outliers. This paper presents a patch based inter-image outlier suppression technique. First the image is divided into square patches. The differences between the patch-wise means of the pre- and intra-opereative images are calculated and represented as the difference of mean(DOM) image. The local spatial gradient of the DOM image is calculated next. A patch is included into outlier region by thresholding the absolute value of the gradient vector. For avoiding the blocking artifacts, an edge refinement procedure follows the said outlier suppression technique. Registration is modelled as a minimization problem of the negative of the sum of square of the normalized cross correlations for all the patches. Particle swarm optimization(PSO) is used as an optimization tool for performing the rigid registration of the pre-operative and the intra-operative US images after outlier suppression. The experimentation has been performed over real patient images and simulated synthetic images. Experimental results show that the proposed outlier rejection method outperforms the existing methods and registration results are also adequate for assisting the surgery effectively.