Mean of Means: An Automatic Liver Segmentation Algorithm
Laramie Paxton, Yufeng Cao, Kevin R. Vixie, Yuan Wang, Chaan S. Ng, Brian Paul Hobbs · 2019
We present an automatic liver segmentation method that utilizes the time series data in conjunction with the BK graph cut algorithm and uses a novel approach of computing a “mean of means” for each of the sample healthy and tumor tissue intensities from a set of different tumors. Thus, there is no training process required since these are computed ahead of time using Regions of Interest provided by radiologists. We also use a Gaussian B-spline to fit these two vector means to curves as an approximation for the intensity signals for the healthy and tumor tissues. This method provides a reasonable degree of accuracy for an automatic segmentation scheme, yielding a mean Dice similarity coefficient (DSC) of 73 percent, a relative volume difference (RVD) of 19.5 percent, and a Jaccard Score (JS) of 58.5 percent. The algorithm is simple to implement computationally, and the mean runtime of 24 seconds is short given that no training process is needed.