Automatic and Robust Estimation of Heart Rate in Zebrafish Larvae

Abstract

Noncontact assessment of heart rate (HR) in Zebrafish larvae, based on a video record of the organism, acquired using a camera mounted on a microscope, has gained enormous significance. Completely automatic and robust estimation of HR from videos of nontransgenic larvae requires the determination of an appropriate region of interest (ROI), followed by suitable signal processing steps. Toward such a goal, we develop a fully automatic and adaptive ROI enclosing a predominant portion of the beating heart, irrespective of the image resolution and zoom. The information within the ROI is used to get one or more time series, to be processed for extracting the signal containing information about the beating heart. Among the various possibilities, we show that the multichannel approach exploiting color information and based on independent component analysis to extract the cardiac signal—is desirable, due to several reasons, including its ability to handle noise, minor movements of the larvae or of the platform, and statistical performance. The proposed sequence of algorithms is validated on videos of 41 larvae (2 days and 4 days postfertilization). The computer estimated values of HR compared well with the ground truth obtained by visual-counting. We have also devised a method of tracking the ROI associated with drifting larvae and tested it on real data. In addition, an example of handling a type of arrhythmia is given. Note to Practitioners —Studying the Zebrafish larva as a model organism, is very useful in developmental biology, toxicology, and pharmacological testing—due to the similarity in the response to various drugs, easiness of breeding, high throughput, lesser cost involved, in vivo testing, and reduced bioethical concerns. The traditional approach of manually counting the heart beat of the larvae kept under a microscope is tedious. Automating the process using videos of the larvae placed under a microscope is therefore important. The existing approaches are either not completely automatic or involve expensive equipment and/or modifying the larvae (biological intervention) such that the heart expresses fluorescence protein, and are not designed to handle even minor motion. We consider completely automating the process of estimating the heart rate from nontransgenic zebrafish larvae—avoiding any kind of manual intervention. The solution to the preceding lies in automating: 1) the determination of the ROI that adapts to the location and size of the larvae, camera zoom, and image resolution; 2) the tracking of the ROI to handle drifting larvae; and 3) estimation of HR through appropriate signal processing, to add to the robustness of the method, including the handling of mild motion. We present a completely automatic and robust HR estimation algorithm, which works automatically to estimate an adaptive ROI, followed by independent component analysis. The method works even in the presence of minor motion of the larvae, including drift. We plan to extend this paper to studying heart rate variability—an important method of quantifying the neural control of the cardiovascular system as well as identifying several pathologies—and certain types of arrhythmia.