Deep Convolutional Neural Networks for left ventricle segmentation.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Cardiology

Background:

• Left ventricle (LV) segmentation is vital for cardiac function analysis.
• Manual segmentation is time-consuming and limited to specific cardiac phases.
• Current semi-automated methods often require manual initialization, introducing errors.

Purpose of the Study:

• To develop a fully automated LV segmentation method using a Deep Convolutional Neural Network (DCNN).
• To compare the performance of DCNN initialization with Gabor filters versus random filters.
• To expedite cardiac morphology analysis and reduce subjectivity in patient care.

Main Methods:

• A Deep Convolutional Neural Network (DCNN) was employed for automatic LV wall segmentation in cardiac MRI images.
• The algorithm calculates pixel probabilities for LV wall or background without manual input.
• Performance was evaluated by comparing DCNN with Gabor filter initialization against random filter initialization.

Main Results:

• The Gabor filter-initialized DCNN achieved an accuracy of 0.97.
• Specificity was 0.984, sensitivity was 0.841, and mean accuracy was 0.902.
• Gabor filters demonstrated superior performance compared to random filters for DCNN-based LV segmentation.

Conclusions:

• Fully automated LV segmentation is feasible using DCNN with Gabor filter initialization.
• This approach enhances the speed and objectivity of cardiac function analysis.
• The findings suggest Gabor filters are effective for improving DCNN performance in cardiac MRI segmentation.