Comparação de técnicas 2D e 3D para segmentação de fígado em tomografia computadorizada

Abstract

Due to the high mortality rate from liver diseases, such as liver cancers, and the complexity in treatment of information in CT scans, responsible for the early detection and diagnosis of liver cancer. It becomes apparent the need for systems CAD (Computer-Aided Diagnosis) as a second opinion in diagnosis, aiming to increase the accuracy in the detection of liver cancer by means of CT. One of the main steps in CAD systems is the segmentation of the object of interest. Thus, many researchers have proposed new techniques of liver segmentation. In this work, we objectively evaluate the performance of two techniques presented in the literature with the objective of comparing and discussing the methodological aspects. The first technique evaluated preprocesses images for noise reduction and contrast enhancement, followed by a deterministic seed launch, and two-dimensional region growing with post-processing. On the other hand, the second technique does not use any pre- or post-processing, realizing the seed launch in a non-deterministic way and a three-dimensional region growing through the response of a one-class classifier. In our evaluation, the first technique obtained a performance of 94.14% while the second technique was 67.54%. Thus, considering the Hit Rate in percentage and the Mean Square Error in cm³, we can conclude that the first technique had a better performance in comparison to the second technique.