Exploring universal segmentation models for automatic quantification of cardiac functional parameters from zebrafish...

Area of Science:

• Cardiovascular toxicology
• Zebrafish model systems
• Image segmentation and analysis

Background:

• Accurate quantification of cardiac function is vital for assessing environmental chemical toxicity.
• Current methods for zebrafish cardiac function evaluation are often manual, time-consuming, and inaccurate.

Purpose of the Study:

• To develop and validate an automated framework for quantifying zebrafish cardiac functional parameters using deep learning.
• To benchmark universal segmentation models for accurate segmentation of zebrafish ventricles and pericardia.

Main Methods:

• Benchmarking 20 state-of-the-art deep segmentation models.
• Utilizing the best-performing model, Mask2Former, for automated segmentation of ventricles and pericardia in zebrafish heartbeat videos.
• Computing seven cardiac functional parameters based on segmented ventricular and pericardial morphologies.

Main Results:

• Mask2Former achieved high segmentation performance: 93.46% IoU and 96.58% Dice for ventricles; 83.31% IoU and 90.89% Dice for pericardia.
• Automatically quantified cardiac parameters demonstrated high accuracy with <10.0% relative error compared to manual measurements.
• The framework successfully quantified heart rate, stroke volume, cardiac output, ejection fraction, and other parameters.

Conclusions:

• The proposed automated framework offers a novel, rapid, and reliable tool for evaluating cardiovascular toxicity of environmental chemicals.
• This approach overcomes limitations of manual and semi-automatic methods, enabling comprehensive functional evaluation.
• The study highlights the potential of deep learning for advancing toxicological assessments in aquatic models.