Objective:This paper proposed an improved K-means clustering algorithm based on kernel density estimation which is used for the automatic segmentation of whole-body bone scan image. Methods:First,we sharpened and smoothed the 2D SPECT whole-body scan image for preprocessing. Second,we used Gaussian kernel density estimation curve to obtain optimal clustering centers as the initial value of K-means clustering algorithm. And then,we segmented the image using the K-means clustering algorithm. Finally,the template match method was performed to delete wrong recognized areas. Results:Image preprocessing provided clearer and more detailed activity structures and improve image quality. The improved K-means clustering algorithm generated a higher degree of Tanimoto similarity than traditional K-means method,and it had a less running time than others. Conclusion:Kernel density estimation can effectively avoid the blindness of the initial clustering center selection in the K-means method,and make the clustering results more rapid,accurate and stable. The proposed algorithm is suitable for whole-body bone scan image segmentation which has important significance to the analysis of region of interest.