Extended Blind End-member and Abundance Extraction for Biomedical Imaging Applications.
D U Campos-Delgado1, O Gutierrez-Navarro2, J J Rico-Jimenez3
1Faculty of Sciences, Universidad Autonoma de San Luis Potosi, SLP, México.
IEEE Access : Practical Innovations, Open Solutions
|January 17, 2020
Summary
The extended blind end-member and abundance extraction (EBEAE) method accurately unmixes biomedical images. This approach enables quantitative analysis of samples with minimal prior information, advancing optical imaging applications.
Area of Science:
- Biomedical Imaging
- Optical Spectroscopy
- Computational Imaging
Background:
- Linear mixture models are crucial in biomedical imaging for identifying image components and their proportions.
- Blind linear unmixing (BLU) aims to determine these components and proportions without prior knowledge, a challenging but vital task.
Purpose of the Study:
- To introduce and validate the extended blind end-member and abundance extraction (EBEAE) methodology for solving the BLU problem.
- To demonstrate the efficacy of EBEAE in diverse biomedical imaging applications.
Main Methods:
- Developed EBEAE based on constrained quadratic optimization and an alternated least-squares strategy.
- Employed a local approach for abundance estimation (entropy maximization) and a global technique for end-member identification (reducing similarity).
- Validated using synthetic datasets with varying noise levels and compared against existing BLU algorithms.
Main Results:
- EBEAE successfully performed blind linear unmixing on synthetic data across different noise conditions.
- Applied EBEAE to multi-photon fluorescence lifetime imaging microscopy (m-FLIM), optical coherence tomography (OCT), and hyperspectral imaging.
- Achieved quantitative analysis in oral cavity, artery, and brain tissue samples, including tumor identification.
Conclusions:
- EBEAE provides a robust and effective solution for blind linear unmixing in optical measurements.
- The methodology demonstrates significant potential for quantitative analysis in various biomedical imaging modalities with minimal a priori information.
