Automatic Left Ventricle Segmentation from Short-Axis Cardiac MRI Images Based on Fully Convolutional Neural Network

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Cardiology

Background:

• Accurate left ventricle (LV) segmentation in cardiac MRI is crucial for diagnosing cardiovascular diseases and assessing cardiac function.
• Manual LV segmentation is time-consuming and labor-intensive for medical experts.
• Automated LV segmentation methods are needed to improve efficiency in clinical practice.

Purpose of the Study:

• To develop and evaluate a fully convolutional network (FCN) for automated LV segmentation in short-axis cardiac MRI.
• To compare the proposed FCN architecture with the U-Net model for LV segmentation performance.
• To enhance segmentation accuracy by addressing class imbalance and applying effective image preprocessing techniques.

Main Methods:

• A fully convolutional network (FCN) architecture was proposed for automatic LV segmentation.
• Experiments compared FCN with U-Net, optimizing hyperparameters like learning rate and batch size.
• Class weighting and pixel normalization were employed to improve feature representation and handle pixel imbalance.

Main Results:

• The proposed FCN achieved high performance metrics: Dice (0.93), Jaccard (0.87), sensitivity (0.98), and specificity (0.94).
• The developed FCN model demonstrated superior segmentation performance compared to the standard U-Net model.
• The results indicate an advanced fully automated method for LV segmentation.

Conclusions:

• The proposed automated LV segmentation method is effective and suitable for clinical application.
• This technique can assist clinicians in diagnosing cardiac diseases more efficiently using short-axis MRI.
• The study highlights the potential of deep learning in improving cardiovascular diagnostics.