Automated Brain Segmentation for Guidance of Ultrasonic Transcranial Tissue Pulsatility Image Analysis

Background and Objective: Tissue pulsatility imaging is an ultrasonic technique that can be used to map regional changes in blood flow in the brain. Classification of regional differences in pulsatility signals can be optimized by restricting the analysis to brain tissue. For 2D transcranial ultrasound imaging, we have implemented an automated image analysis procedure to specify a region of interest in the field of view that corresponds to brain. Methods: Our segmentation method applies an initial K-means clustering algorithm that incorporates both echo strength and tissue displacement to identify skull in ultrasound brain scans. The clustering step is followed by processing steps that use knowledge of the scan format and anatomy to create an image mask that designates brain tissue. Brain regions were extracted from the ultrasound data using different numbers of K-means clusters and multiple combinations of ultrasound data. Masks generated from ultrasound data were compared with reference masks derived from Computed Tomography (CT) data. Results: A segmentation algorithm based on ultrasound intensity with two K-means clusters achieves an accuracy better than 80% match with the CT data. Some improvement in the match is found with an algorithm that uses ultrasound intensity and displacement data, three K-means clusters, and addition of an algorithm to identify shallow sources of ultrasound shadowing. Conclusions: Several segmentation algorithms achieve a match of over 80% between the ultrasound and Computed Tomography brain masks. A final tradeoff can be made between processing complexity and the best match of the two data sets.