An Evaluation Study of the CAD Segmenting Algorithm Based on the ROC Analysis and ALVIM Phantom
Xue Wei-jing · Zhongguo yixue wulixue zazhi · 2007
Objective: This essay targets to compare various segmenting algorithm systems by ROC analysis and ALVIM Model phantom. It aims to get the comparative outcome of the accuracy of the segmenting algorithm and the inspecting capacity of signals and finally suggests a method that is appropriate for the horizontal comparison among segmenting algorithm systems. Methods: By drawing up four integral mammary gland segmenting algorithm systems, the author will segment the actual phantoms and the ALVIM phantoms, calculating the overlapping ratio to the gold standard and the signals' inspecting capacity parameter for segmenting true positive ratio and false negative ratio by applying the ALVIM phantom statistics; then the author will identify the model phantom with the help of the mentioned segmenting contour, analyze the ROC curve and eventually find out the efficiency of the four segmenting algorithm systems in the identifying process. Results: This essay will gain the overlapping ratio between the outcome by segmenting algorithm systems and the artificial segmenting area, the signals' inspecting capacity by each segmenting algorithm systems and the efficiency of the segmenting algorithm helping diagnosing doctors read the X-ray images. Conclusions: The ROC analysis and segmenting ALVIM phantom can fully examine the segmenting effects of the segmenting algorithm systems and the signals' inspecting capacity, horizontally comparing the different segmenting algorithm systems and testing the accuracy and robust of the systems. The algorithm applying five factors' determining method is simple and appropriate; hence the ROC comparative analysis between large data and distinct methods becomes convenient.