2-dimensional fusion of cerebral cross-modality images employing a mutual information algorithm*
C. Kollmann †,1 , B. Greiffenberg 2 , F. Schlachetzki 2 , U. Bogdahn 2 , H. Bergmann 1
1. Department of Biomedical Engineering
& Physics, Univ. Vienna, Waehringer Guertel 18-20, A-1090
Vienna (Austria)
2. Department of Neurology, University of Regensburg,
Bezirksklinikum Regensburg, Universitaetsstr. 84, D-93053
Regensburg (Germany)
Manuscript received: April 27, 2001; revised July 17, 2001
Accepted for publication: July 24, 2001
Abstract
Diagnosis and treatment monitoring of neurological diseases require a variety of different functional and anatomical neuroimaging procedures. However, each of these favour or lack specific bio-physical information e.g., on the cerebral parenchyma, and neurologic disease requiring complex interpretation by the physician. Image fusion may be a suitable solution to gather different information of the brain in one image and, thus, enable a more accurate diagnosis. In this pilot study, 2-dimensional (2D) cranial computed tomography (CCT), magnetic resonance tomography (MRT) and three-dimensional (3D) transcranial color-coded sonography (TCCS) data sets were registered and fused with the ANALYZE-AVW software (Biomedical Imaging Resource, Mayo Foundation, Rochester, USA). A procedure was developed allowing rapid overlay of the images. First, identical anatomic structures in each data set were identified and segmented before a mutual information algorithm was used to create a transformation matrix. With the knowledge of this matrix one of the different modalities could be registered to the other modality. In a final step, fusion of the two image modalities was performed. 2D image registration and fusing of CCT / MRT with TCCS was achieved in a short time resulting in images presenting multiple pathological features of various neurologic diseases. Additional information on brain structures as well as flow data in cerebral vessels as detected by ultrasound were overlaid to CCT and MRT images with high accuracy. Image fusion may be a potential solution to enhance modern neuroimaging tools. Further studies have to be pursued focusing on the following questions: the stability and accuracy of the mutual information algorithm for fusion of 3D data sets, and the optimal intensity and color map for each image data set.
KEYWORDS: Multimodality, mutual information, neuroimaging, ultrasound.