Comparison between elliptical and squared ROI to launch an automatic seed to region growing algorithm on hepatic segmentation using CT images

Abstract

Early detection of liver cancer increases the survival rate by 60-70% in five years. As part of CAD development for the liver, two automatic seed selection techniques (for automatic liver segmentation using region growing technique) are compared, using the calculation of Mean Squared Error (MSE) to analyze which one has the lowest error when compared with manual segmentation. Was used 2631 liver computer tomography slices, the first step is make the pre-process of these slices with the windowing and then, is applied a mean filter with the gamma transformation. The first technique defines the Seed Launch on Elliptical Area (SLEA) for find the seed point, and the second technique defines the Seed Launch into Square Area (SLSA) for find the seed point. And the seed points that were found with each one of these techniques are used to perform the liver segmentation with the region growing algorithm, and from the liver segmented is performed the entire exam volume and the MSE calculation for each technique, and these results compared with the manual segmentation. The average liver volume measured manually was of 1,388.41 cm3, and with SLEA technique the average liver volume measured automatically was of 1.663,59 cm3 and the average MSE was of 68.69±96.35 cm3, and for the SLSA technique the average liver volume measured automatically was of 1,661.43 cm3 and the average MSE was of 193.16±286.84 cm3. The smallest MSE SLEA was of 2.035 cm3 and the bigger was of 391.815 cm3, and for the SLSA the smallest MSE was of 2.122 cm3 and the bigger was of 1,110.824 cm3. Thus was observed that both techniques have the average volume similar, but the SLEA technique has the smallest values and averages of the MSE.