Automatic Extraction of the Central Symmetry (Mid-Sagittal) Plane from Neuroradiology Images

Normal human brains present an approximate bilateral symmetry. This symmetry is reflected in CT and MR images depicting axial and coronal slices of the brain. Though the internal structure of a pathologic brain may depart from its normal bilateral symmetry, the ideal imaginary bilateral symmetry plane remains invariant. This plane is often referred to as the mid-sagittal plane of the brain. Automatically detecting and explicitly representing the mid-sagittal plane can benefit image understanding in neuroradiology in many ways, including registration, lesion detection, screening and diagnosis. We have developed a simple yet effective algorithm for extracting the axis of bilateral symmetry from each axial brain slice and combining results from multiple slices to determine the central plane of bilateral symmetry for the 3D head. This algorithm has been tested on 12 sets of CT and MR normal and pathological neural scans – a total of 400+ single images. Given a set of axial scans, the accuracy of the algorithm is within one degrees in terms of yaw (< 0:3 degrees) and roll (< 0:75 degrees) angles, and under five pixels in terms of the offset of the symmetry axis. These results are used to correct roll and yaw rotational errors in input images, to automatically produce the midsagittal plane of a given set of axial brain scans, and to detect asymmetries that may be caused by lesions or mass effect in the brain without human intervention.