Phasor-FLIM (Fluorescence Lifetime Imaging Microscopy) is an emerging non-invasive fluorescence based technique which is capable of analyzing the metabolic profile of cells through detection of autofluorescent NADH. It has been shown that Phasor-FLIM is effective for quickly identifying the effectiveness of different drugs on different cell lines, which suggests that it can be a powerful tool for precision medicine. Still, there are some aspects of Phasor-FLIM that can be improved. While images are rapidly acquired on the scale of one second per frame, other aspects of experimentation are monotonous and time-consuming, data analysis specifically. Data analysis involves manually removing background artifacts from the image and in certain cases, compartmentalizing the nucleus and cytoplasm via mouse drawing. We propose incorporating machine learning algorithms to automate data analysis and reduce human bias in background removal and compartmentalization. We train our algorithm using images expressed as matrices of pixel intensities and frequencies of emission. After training, we validate and test the algorithm’s ability to evaluate differences between the background and cells imaged with the aforementioned two characteristics. The implementation of machine learning in Phasor-FLIM will greatly reduce time spent analyzing data and increase the speed at which drug efficacy can be determined on patient tissues.




Author: Justin Wang

Status: Project Concept