Brain Tumor Segmentation Based on Structuring Element Map Modification and Marker-controlled Watershed Transform

Brain medical images are generally prone to noise and also fraught with intensity heterogeneity within the tumor. Fuzzy and boundary discontinuity caused by the tumor also adversely affects the accuracy of the tumor segmentation. A method based on morphological structuring element map modification and markercontrolled watershed segmentation is proposed. Firstly, a structuring element map is constructed according to the sum of the weighted variance of the specific regions within morphological gradient image, and each value of the structuring element map represents the size of structuring element (SE). Secondly, the original image is modified by morphological opening-closing, where the size of SE are determined by the structuring element map in the corresponding pixel, such an adaptive image modification can eliminate the noise and small regular details while preserve the larger object contours without less location offsets. Finally, marker-controlled watershed transform is used to complete the tumor segmentation. Experiments show that the method ensures brain tumors are more accurately segmented.