Investigating Metastasis Based on CT-Scan Tumor Edge Detection

摘要

The liver is necessary for survival and is also prone to many diseases. The CT examinations can be used to plan and properly administer radiation treatments for tumors and to guide biopsies and other minimally invasive procedure. Manual segmentation and classification of CT image is a tedious task and time consuming process which is impractical for large amount of data. Fully automatic and unsupervised methods eliminate the need for manual interaction. In this paper, we propose a histogram based method to detect metastasis liver abnormality in CT images automatically. This proposed method is composed of three steps. In the first step, all kind of noises are removed such as speckles using image filtering. The overlap between different peaks is an strong evidence of noisy image. In the second step, metastasis tumor candidates are detected using histogram based analysis and K-mean based analysis. Applying histogram based analysis algorithm leads to remove the overlap between liver and the tumor. Suspected area was recognized successfully as the outcome of histogram based analysis. Tumor Pattern shows gradual change from dark to light. The darker tune means worst damage as well as older than the lighter tune. The dark tune indicates severity and old. The light tune indicates new development of the tumor. Quantitative evaluation was done using ANOVA single factor test analysis to test whether there is any significant relation between the classes. Since P < 0.05, there is insignificant relation between all the classes and we reject the null hypothesis. Further, validation between manual and automated segmentation was made and found that the error between manual segmentation and automated segmentation is smaller than 1% which shows an evidence of success. In the final step, the performance capability of K-means versus HBAA was made. The error percentage in (HBAA) is (3.9%), while in (K-mean classifier) the 35.6%. The estimated area by (K-mean classifier) was exaggerated more than one third. The estimated area by (HBAA) was 96% of the calculated area by the radiologist. The result is a proof of the superiority of (HBAA) over (K-mean classifier).