A data augmentation approach to train fully convolutional networks for left ventricle segmentation.

Area of Science:

• Medical imaging analysis
• Cardiovascular disease diagnostics
• Machine learning in healthcare

Background:

• Left ventricle (LV) segmentation is crucial for assessing cardiac contractile function and diagnosing cardiovascular diseases.
• Fully convolutional networks (FCNs) are effective for image segmentation but require substantial annotated data, posing a challenge for LV segmentation due to limited sample availability.
• Overfitting is a significant concern when training FCNs with small datasets.

Purpose of the Study:

• To investigate the impact of data augmentation on FCN performance for LV segmentation.
• To propose a novel data augmentation technique utilizing shape models to train FCNs with limited annotated samples.
• To enhance the accuracy and reliability of automated LV segmentation.

Main Methods:

• Analysis of the influence of training sample augmentation on FCNs for LV segmentation.
• Development and application of a data augmentation approach based on shape models.
• Training and evaluation of FCNs using augmented datasets on four public datasets.

Main Results:

• Balanced training samples significantly impact FCN performance in LV segmentation.
• The proposed shape model-based data augmentation approach enables effective FCN training with few samples.
• The FCN trained with augmented data demonstrated superior performance compared to existing automated segmentation methods.

Conclusions:

• Data augmentation, particularly using shape models, is a viable strategy to overcome data limitations in FCN-based LV segmentation.
• The developed method improves the accuracy of cardiac segmentation, aiding in cardiovascular disease diagnosis.
• This approach offers a promising solution for robust and efficient automated LV segmentation in clinical settings.