UC Irvine

Zebrafish, revered as an invaluable model organism, is widely employed in cardiovascular disease research. It facilitates a comprehensive understanding of cardiac behaviors and conditions by appraising cardiac functions extracted from heartbeat echo-videos. Researchers often scrutinize the ejection fraction, a metric indicative of heart performance. Nevertheless, current techniques for such evaluations grapple with numerous challenges. These methods are laborious, time-consuming, and prone to errors, making them unfavorable for large-scale investigations. These limitations are particularly troublesome when dealing with massive datasets or when detailed assessments are required, hindering their effectiveness in applications like high-throughput screening for drug discovery. Addressing these constraints, an approach was conceived to enhance cardiac function analysis in zebrafish. This method utilizes a deep learning model to enable the automatic evaluation of ejection fractions from heart echo-videos. This thesis outlines an approach hinging on a specific deep learning model architecture. The model's accuracy, confirmed using the Dice coefficient and the Intersection over Union (IoU) score, stood at a robust 0.967 and a significant 0.937, respectively. The testing phase yielded a promising error rate range from 0.11% to 16.96%, averaging 5.13%, attesting to the method's accuracy and reliability. Furthermore, this method can be assimilated into existing lab settings, synergizing with binary recordings to optimize large-scale video analysis and improve high-throughput screenings' efficacy. Compared to traditional techniques, this deep learning-based method simplifies zebrafish cardiac function monitoring and quantification, thus boosting laboratory efficiency. In conclusion, this approach heralds a notable advancement in zebrafish cardiovascular research. Enhancing the speed, accuracy, and ease of cardiac function analysis holds significant potential to transform the study of cardiovascular diseases, serving as a crucial tool for researchers in this pivotal field.