Related Experiment Video
Updated: May 2, 2026
10:39
A Label-Free Segmentation Approach for Intravital Imaging of Mammary Tumor Microenvironment
Published on: May 24, 2022
2.0K
Variable importance in nonlinear kernels (VINK): classification of digitized histopathology
Shoshana Ginsburg1, Sahirzeeshan Ali1, George Lee2
1Department of Biomedical Engineering, Case Western Reserve University, USA.
Medical Image Computing and Computer-Assisted Intervention : MICCAI ... International Conference on Medical Image Computing and Computer-Assisted Intervention
|March 1, 2014
Summary
This study introduces Variable Importance in Nonlinear Kernels (VINK), a new method for feature selection in digital pathology. VINK improves classification and regression performance using nonlinear dimensionality reduction techniques.
Area of Science:
- Digital pathology
- Computational biology
- Medical image analysis
Background:
- Quantitative histomorphometry uses image features for disease modeling, but high dimensionality poses challenges.
- Traditional dimensionality reduction (DR) methods limit feature interpretability.
- Existing variable selection methods are not suitable for nonlinear DR (NLDR).
Purpose of the Study:
- To develop a method for quantifying variable importance in nonlinear kernel embeddings.
- To enable feature selection for NLDR techniques in digital pathology.
Main Methods:
- Developed Variable Importance in Nonlinear Kernels (VINK) to approximate the mapping between original and kernel PCA (KPCA) feature spaces.
- Integrated VINK with Isomap and Laplacian eigenmap algorithms.
- Evaluated VINK on three digital pathology tasks.
Main Results:
- VINK successfully quantified variable importance in nonlinear kernel embeddings.
- Feature subsets identified by VINK achieved comparable or superior performance to original high-dimensional sets.
- Demonstrated VINK's utility in predicting prostate cancer failure, breast cancer recurrence risk, and oropharyngeal tumor outcomes.
Conclusions:
- VINK offers a robust approach for feature selection in NLDR for digital pathology.
- This method enhances the interpretability and performance of machine learning models in medical image analysis.
- VINK facilitates more effective use of complex histopathology data for clinical prediction.
