Segmentation of humeral head from MR slices

Abstract

The purpose of this study is to segment the humeral head from two dimentional proton dansity(PD) magnetic resonance images (MRI) of shoulder. MRI studies of patients presenting with shoulder pain may demonstrate bone marrow edema and different degrees of contour changes. Computerised tomography(CT) and T1 weighted MR images demonstrate bone margins better. T2 or PD weighted images are poor in determining exact images of bone margins, instead they give better detection of edema and thus pathology. However PD weighted slices have a higher noise to signal ratio than other modalities. Moreover soft tissues in close proximity to humeral head which have the similar intensity with bone make the segmentation of bony humeral head a challenging problem in PD weighted images. In this study we used a new data set which consists of PD weighted axial images of 33 shoulder MRI examinations of 33 randomly selected patients. We used specle reducing anisotropic diffusion(SRAD) method to decrease noise in PD weigted images. Secondly using Hough transform the round shaped humeral head was located roughly, afterwards for a more successful segmentation we applied active contour without edge (ACWE) method. Success of these methods is determined by comparing our results with manual segmentaion results of an expert and evaluated by Sorensen-Dice metric. The average value of measure over the entire database is found to be 90% by ACWE method. Application of ACWE method on PD weighted images provided higly successfull results for segmentation of bony structures. There is no other study to de determine bone contours in PD images in the literature.