Files
Abstract
Intraflagellar transport (IFT) is essential for the construction of cilia and flagella in eukaryotic cells and plays a critical role in the transport of protein cargo to and from the cell body. Protein molecules involved in IFT are observed to travel at different velocities. These velocities are computed using spatio-temporal maps called kymographs. Kymographs are single images that represent the 3D microscopy images of intracellular motion as 2D time series data. In vivo microscopy imaging typically results in the generation of noisy kymographs which are difficult to analyze. Existing techniques for IFT velocity measurement entail manual detection of IFT trails on the kymographs, followed by computation of the IFT velocities via determination the slope of each IFT trail. Since manual kymograph analysis is laborious, time consuming and error prone, an automated algorithm to extract IFT trails and determine the IFT velocities in kymographs with minimal manual intervention is a valuable tool for biologists. To this end, we propose a machine learning-based approach to the analysis of kymographs that segments and delineates the IFT trails, and computes the IFT protein velocities. In the proposed approach, the kymograph image is preprocessed to suppress noise and identify potential IFT trail pixels. The potential IFT trail pixels are further characterized using Gabor wavelet transform (GWT) and curvelet transform (CT) features. A Support Vector Machine (SVM) is used to classify the potential IFT trail pixels into three IFT trail categories followed by the extraction of continuous IFT trajectories and the computation of the IFT velocity associated with each IFT trajectory. Experimental results show the advantages of the proposed approach in terms of accuracy and greatly reduced processing time for kymograph analysis.