A learning-based automatic segmentation and quantification method on left ventricle in gated myocardial perfusion...

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Cardiovascular Imaging

Background:

• Accurate left ventricular (LV) functional assessment in gated myocardial perfusion SPECT (MPS) relies heavily on precise myocardium segmentation.
• Current manual segmentation methods are time-consuming, subjective, and prone to variability.
• There is a need for automated, objective, and efficient LV segmentation techniques in gated MPS.

Purpose of the Study:

• To develop and validate a novel machine learning-based method for automated LV myocardium segmentation in gated MPS imaging.
• To accurately measure LV myocardial volume without human intervention.
• To improve the efficiency and objectivity of LV functional assessment in gated MPS.

Main Methods:

• An end-to-end fully convolutional neural network was employed for LV myocardium segmentation, delineating both endocardial and epicardial surfaces.
• A novel compound loss function was utilized during training to enhance prediction accuracy by promoting similarity and penalizing discrepancies.
• The method was retrospectively validated on 56 patients (32 normal, 24 abnormal), comparing automated segmentation contours with physician-delineated ground truth.

Main Results:

• The automated segmentation demonstrated excellent agreement with physician-drawn contours, achieving average Dice Similarity Coefficient (DSC) metrics > 0.900 and Hausdorff distance < 1 cm across all phases and patients.
• A high correlation coefficient (0.910 ± 0.061, P < 0.001) was observed for LV myocardium volume between the automated method and ground truth.
• The mean relative error for LV myocardium volume was low (-1.09 ± 3.66%), indicating high accuracy.

Conclusions:

• The developed machine learning method is feasible and accurate for quantifying LV myocardium volume changes throughout the cardiac cycle.
• Automated segmentation of LV myocardium in gated MPS imaging using this learning-based approach shows significant potential for clinical application.
• This technology promises to enhance the reliability and efficiency of LV functional assessment in nuclear cardiology.