Motivation: Deep learning offers potential for automated cardiac segmentation in preclinical models but remains challenging due to limited rodent CMR data and high imaging variability. Goal(s): To develop an accurate, fully automated LV segmentation deep learning model to reduce analysis time and manual workload in preclinical cardiac assessments. Approach: Cine SAX CMR data from 73 mice were segmented at systolic and diastolic phases, producing 910 annotated images. A 2D U-Net model optimised for small, imbalanced datasets was trained and validated using DICE similarity and cardiac volume metrics. Results: The model achieved high accuracy and speed, enabling reliable LV assessments and supporting larger preclinical studies. Impact: Automating left ventricular segmentation with UNet3Plus has the potential to reduce analysis time and human error in large-scale, multi-timepoint preclinical studies, improving efficiency and enabling deeper investigation into cardiac dysfunction to accelerate the development and evaluation of novel cardiovascular therapies.
Hanisah et al. (Tue,) studied this question.