Abstract
3-D liver segmentation is vital in computer-assisted surgery applications such as minimal invasive surgery, targeted drug delivery,tumor resection, and donor transplantation. This study describes development and evaluation of a novel liver segmentation paradigm in support of Selective Internal Radiation Therapy (SIRT). Since segmentation accuracy and computational simplicity are the two key features for evaluation, the proposed method couples a modified k-means based segmentation and localized-contouring algorithm to obtain segmentation with high accuracy, based on an optimal number of slices. Furthermore, parallel computing is used to reduce the high computational load required of the process. Minimal manual interaction was required in the form of initialization with no correction or adjustment done during or after the process completion. Five rounds of experiments were performed to determine the accuracy and computational performance of the segmentation algorithm. Results were assessed by comparing volumes obtained from the segmentation algorithm to those obtained by manual segmentation done by experts. Statistical analysis is also carried out to determine if the same accuracy is obtained during multiple runs of the dataset and to determine if the manual initialization has any impact on the accuracy of the results. An average accuracy of 98.27% was achieved in estimating the liver volumes with consistent results obtained in various runs and independently of the user initializing the task. A reduction of 78% in computational time was accomplished by the parallel computing techniques in support of the lengthy segmentation process. Since SIRT requires accurate calculation of the liver volume, this new method provides highly accurate and computationally efficient process required of such challenging clinical requirements.
