Open Implementation of Feature Extraction Methods for Computer Aided Polyp Detection with Principle Component Analysis
Abstract
Computer-aided polyp detection (CAPD) automatically detects colonic polyps and presents them to the user in either a first or second reader paradigm, with the goal of reducing examination time while increasing detection sensitivity. There has been a tremendous amount of research on CAPD, however a lack of source code availability in commercial applications impairs the comparison of features and inhibits effective research practices due to the limited space available to fully describe methods in publications. In this paper, we survey the commonly used feature extraction methods, describe their implementation, and compare the features using principle component analysis(PCA). A total number of 89 features have been calculated and reduced to 10 based on PCA. The result shows that in true positive(TP) class, Gaussian and principal curvature carry more information than other features in indentifying the poly candidates. On the other hand, false positive(FP) class shows an abundant variability in their shape. Although, Gaussian and principal curvature has relatively heavy weight, they are not as dominant as in TPs class.
Reference
Yuan Shen and Christopher L. Wyatt, Open Implementation of Feature Extraction Methods for Computer Aided Polyp Detection with Principle Component Analysis, MICCAI 2008 Workshop on computational and visualization challenges in virtual colonoscopy. preprint
Code and Data
The code depends on ITK (version 3.4 is known to work). It has been built and tested on Linux-x86 (32 and 64 bit) and requires a machine with at least 2 GB of RAM to run properly. The test takes approximately 50 minutes to run.
Full source code and tests (including example data): download version 1.1 (123 MB tar.gz).
Older Versions: href="FeatureExtraction_1.0.tar.gz">Version 1.0.
Comments and Bug Reports