Frequency-constrained robust principal component analysis: a sparse representation approach to segmentation of dynamic features in optical coherence tomography imaging

James P Mclean,Christine P Hendon,Yuye Ling

doi:10.1364/oe.25.025819

Abstract

Sparse representation theory is an exciting area of research with recent applications in medical imaging and detection, segmentation, and quantitative analysis of biological processes. We present a variant on the robust-principal component analysis (RPCA) algorithm, called frequency constrained RPCA (FC-RPCA), for selectively segmenting dynamic phenomena that exhibit spectra within a user-defined range of frequencies. The algorithm lacks subjective parameter tuning and demonstrates robust segmentation in datasets containing multiple motion sources and high amplitude noise. When tested on 17 ex-vivo, time lapse optical coherence tomography (OCT) B-scans of human ciliated epithelium, segmentation accuracies ranged between 91-99% and consistently out-performed traditional RPCA.

Highlights

Segmentation, detection, tracking, and analysis of motion are challenges in medical imaging that span multiple applications and modalities [1,2,3,4,5]
Data was acquired using a custom Ultra-High Resolution optical coherence tomography (OCT) (UHR-OCT) system designed in house that has previously been used for imaging human trachea samples [22, 24, 25]
The sparse component generated by FC-Robust Principal Component Analysis (RPCA) was shown to effectively rejects pixels associated with spectral features outside the specified frequency constraint such as nearby mucus clouds

Summary

Introduction

Segmentation, detection, tracking, and analysis of motion are challenges in medical imaging that span multiple applications and modalities [1,2,3,4,5]. Understanding dynamic biological phenomenon necessitates quantitative analysis, which relies on tools for separation of the motion of interest from other static components in the scene [6,7]. Such separation of static and dynamic components is in general difficult due to variety in types of motion (particle flow, physical deformation, periodic motion, etc...). Sparse Representation theory offers many favorable tools for solving the kind of dynamic segmentation problems encountered in imaging. Assuming some signal Y ∈ Rm×n contains both static and dynamic components (L, S ∈ Rm×n, respectively), RPCA says that L and S can be estimated from the observation Y by solving the optimization problem: minimize L ∗ + λ S 1

Methods

Results

Discussion

Conclusion