VII.     Conclusion

This paper proposed Accelerated Kernel Feature Analysis (AKFA), a faster and more efficient feature extraction algorithm derived from the Sparse Kernel Feature Analysis (SKFA) as a means for the detection of polyps on CT colonographic images.  The time complexity of AKFA is , which was more efficient than the  time complexity of SKFA and the complexity  of a more systematic principal component analysis (KPCA). The polyp classification experiment using a k-nearest neighbor classifier showed that  Weighted Kernel optimized KPCA and AKFA yields the superior classification performance to that of KPCA and AKFA, thus demonstrating that the features extracted by WKOAKFA are practically useful in discrimination of polyps from false positives detections.  Therefore, WKOAKFA has the potential to lead a model-based CAD scheme yielding high detection performance of polyps. Such a CAD scheme has the potential of making CT a viable option for screening large patient populations, resulting in early detection of colon cancers and leading to reduced mortality due to colon cancer.


This work was supported in part by USPHS Grant CA095279 and American Cancer Society Research Scholar Grant RSG-05-088-01-CCE. The authors would like to acknowledge X. Jiang who programmed the first matlab implementation of AKFA for improving the computation time.  The authors are also grateful to R. Snapp and X. Zhu at the University of Vermont for providing initial feedback on the appropriateness of various strategies for the course of the project.