Left ventricle segmentation in fetal echocardiography using a multi-texture active appearance model based on the...

Area of Science:

• Medical Imaging
• Cardiology
• Image Analysis

Background:

• Fetal echocardiography is crucial for early cardiac defect detection.
• Assessing fetal cardiac function relies on measurements of heart cavity dimensions and shape.
• Accurate segmentation of cardiac structures is vital for reliable analysis.

Purpose of the Study:

• To develop an automatic segmentation method for the left ventricle in fetal echocardiography.
• To improve the analysis of fetal cardiac function through precise left ventricle measurements.
• To address challenges in fetal cardiac ultrasound image analysis, such as variations in orientation and shape.

Main Methods:

• A novel multi-texture active appearance model (AAM) based on the Hermite transform (HT) was designed for left ventricle segmentation.
• The method utilizes steered HT coefficients for efficient texture pattern coding and local orientation analysis.
• An automatic initialization technique combining thresholding, mathematical morphology, and correlation was developed using a pre-segmented image database.

Main Results:

• The algorithm achieved an average Dice coefficient of 0.8631 and a point-to-curve distance of 2.027 pixels on a database of 143 fetal heart images.
• Performance was validated using various metrics and compared favorably against other segmentation methods.
• The method demonstrated robustness in handling variations in orientation and shape typical of fetal cardiac ultrasound images.

Conclusions:

• An automatic algorithm for left ventricle segmentation in fetal echocardiography was successfully developed.
• The proposed multi-texture AAM scheme provides efficient and accurate segmentation of the left ventricular cavity.
• The method effectively addresses common challenges encountered in fetal echocardiography analysis.