Automatic segmentation of intravital fluorescence microscopy images by K-means clustering of FLIM phasors
Abstract
Fluorescence lifetime imaging microscopy (FLIM) provides additional contrast for fluorophores with overlapping emission spectra. The phasor approach to FLIM greatly reduces the complexity of FLIM analysis and enables a useful image segmentation technique by selecting adjacent phasor points and labeling their corresponding pixels with different colors. This phasor labeling process, however, is empirical and could lead to biased results. In this Letter, we present a novel and unbiased approach to automate the phasor labeling process using an unsupervised machine learning technique, i.e., K-means clustering. In addition, we provide an open-source, user-friendly program that enables users to easily employ the proposed approach. We demonstrate successful image segmentation on 2D and 3D FLIM images of fixed cells and living animals acquired with two different FLIM systems. Finally, we evaluate how different parameters affect the segmentation result and provide a guideline for users to achieve optimal performance.
Yide Zhang
NIH K99 Postdoctoral Fellow
My research is interdisciplinary and focused on developing new types of optical imaging techniques that could advance the work of other researchers and medical personnel in a wide variety of fields. Currently, I am developing next-generation photoacoustic and ultrafast imaging techniques that can observe biological and physical phenomena that are too fast to be imaged with existing methods. The observation of the ultrafast phenomena could provide a better understanding of the fundamentals of life and physical sciences. I am also developing novel quantum imaging approaches that can investigate biological organisms with an imaging performance that cannot be achieved using classical optical imaging. In my free time, I enjoy cooking, hiking, cycling, and traveling.