Automatic Bone Segmentation by a Gaussian Modeled Threshold

摘要

This work presents a new method for automatic threshold selection and its application for bone segmentation in CT images. Based on the mean (μ) and standard deviation (σ) values of an automatically selected region from a Gabor filter response, the proposed method prevents the misclassification of medium and high-valued pixels in images with high density of low-valued (background) pixels like those in medical images. The method obtains an average accuracy of 98.9% and a mean local accuracy of 76.7% using a database of 60 CT images. In addition, the proposed method shows a better performance than the comparative threshold (Otsu and Kittler) and clustering (Fuzzy C-means and K-means) methods applied under same conditions.