Background  Echocardiography is one of the most commonly used and economical radiological examination methods in clinical practice, but there are still many difficulties in the diagnosis of echocardiographic images, such as the low quality of echocardiographic images, significant subjective differences, and dependence on the experience of senior physicians. The ultrasound imaging intelligent diagnosis system based on artificial intelligence (AI) technology is an important support for AI model training based on high-quality ultrasound imaging delineation data.

  Objective  To investigate the “subjective difference” in left ventricular endocardial border delineation by echocardiography, and to perform an objective quantitative evaluation.

  Methods  Echocardiographic apical two-chamber (A2C) and apical four-chamber (A4C) videos were collected from 442 patients in the Chinese PLA General Hospital database who were treated from June to August in 2021. Three senior echocardiographic physicians from grade A tertiary hospitals were invited to select the end-diastolic and end-systolic images and perform left ventricular endocardial border delineation to form a reference standard, then four medical imaging analysts performed double-blind delineation of the end-diastolic and end-systolic left ventricular endocardial border, and finally the quality of delineation by medical imaging analysts (measure the difference between their delineation results and the reference standard of senior physicians) was quantified by calculating the Dice index for contour similarity and the difference of left ventricular ejection fraction (LVEF) (ΔEF). According to image quality, the echocardiographic videos were divided into good-, medium-, and poor-quality groups, and the quality of delineation by medical imaging analysts was compared after two training sessions to quantitatively analyze the efficacy of echocardiographic delineation training.

  Results  The Dice index for the quality of left ventricular endocardial border delineation by medical imaging analysts decreased with the reduction in echocardiographic image quality, with a greater influence on end-diastolic delineation than end-systolic delineation. Median ΔEF for medical imaging analysts was generally a negative value, indicating that the EF value of medical imaging analysts was lower than the reference standard EF value of senior physicians. Contour delineation retraining improved ΔEF between all four medical imaging analysts and the reference standard, and median ΔEF was increased by 3.5%-6.0%.

  Conclusion  The quality of echocardiographic images has a significant influence on the quality of left ventricular endocardial border delineation by medical imaging analysts, and retraining can reduce the difference between their results and the reference standard.