Consensus-Locally Linear Embedding (C-LLE): Application to Prostate Cancer Detection on Magnetic Resonance Spectroscopy

Authors:
Pallavi Tiwari;Mark Rosen;Anant Madabhushi
Affiliations:
Department of Biomedical Engineering, Rutgers University, USA;Department of Radiology, University of Pennsylvania, USA;Department of Biomedical Engineering, Rutgers University, USA
Venue:
MICCAI '08 Proceedings of the 11th International Conference on Medical Image Computing and Computer-Assisted Intervention, Part II
Year:
2008

Citing 5
Cited 1

Bagging predictors

Machine Learning
Local multidimensional scaling

Neural Networks - 2006 Special issue: Advances in self-organizing maps--WSOM'05
Riemannian Manifold Learning

IEEE Transactions on Pattern Analysis and Machine Intelligence
Investigating the Efficacy of Nonlinear Dimensionality Reduction Schemes in Classifying Gene and Protein Expression Studies

IEEE/ACM Transactions on Computational Biology and Bioinformatics (TCBB)
A hierarchical unsupervised spectral clustering scheme for detection of prostate cancer from magnetic resonance spectroscopy (MRS)

MICCAI'07 Proceedings of the 10th international conference on Medical image computing and computer-assisted intervention

Statistical shape model for manifold regularization: Gleason grading of prostate histology

Computer Vision and Image Understanding

Quantified Score

Hi-index	0.00

Visualization

Abstract

Locally Linear Embedding (LLE) is a widely used non-lineardimensionality reduction (NLDR) method that projectsmulti-dimensional data into a low-dimensional embedding space whileattempting to preserve object adjacencies from the originalhigh-dimensional feature space. A limitation of LLE, however, isthe presence of free parameters, changing the values of which maydramatically change the low dimensional representations of thedata. In this paper, we present a novel Consensus-LLE (C-LLE)scheme which constructs a stable consensus embedding from acrossmultiple low dimensional unstable LLE data representations obtainedby varying the parameter (κ) controlling locallylinearity. The approach is analogous to Breiman's Bagging algorithmfor generating ensemble classifiers by combining multiple weakpredictors into a single predictor. In this paper we demonstratethe utility of C-LLE in creating a low dimensional stablerepresentation of Magnetic Resonance Spectroscopy (MRS) data foridentifying prostate cancer. Results of quantitative evaluationdemonstrate that our C-LLE scheme has higher cancer detectionsensitivity (86.90%) and specificity (85.14%) compared to LLE andother state of the art schemes currently employed for analysis ofMRS data.