Proposing novel methods for simultaneous cardiac cycle phase identification and estimating maximal and minimal left atrial volume (LAV) from apical four-chamber view in 2-D echocardiography

Informatics in Medicine Unlocked

Ali Hosseinsabet¹  Toktam Khatibi²  Niloofar Barzegar³ 

[1] Corresponding author.;School of Industrial and Systems Engineering, Tarbiat Modares University (TMU), Tehran, Iran;School of Industrial and Systems Engineering, Tarbiat Modares University, Iran;

关键词: Medical image analysis;    Cardiovascular image;    Cardiac ultrasound;    Deep neural networks;    Motion history image;   

DOI  :  

来源: DOAJ

【 摘 要 】

1 Abstract: Left atrial volume (LAV) estimation is an important issue for prognosis of some adverse cardiovascular events. Manual estimation of LAV is a tedious and time-consuming labor. LAV measurement is a challenging task due to some factors such as artifacts and speckle noise generated by ultrasound imaging, vague boundaries of anatomical structures, viewpoint variations and different scanning angles. Therefore, using automatic methods for estimating LAV is necessary. In this study, our aim is estimating maximal and minimal LAV from echocardiographic images. Moreover, cardiac cycle phase is identified via recognizing end-systole and end-diastole frames as the main prerequisite of LAV measurement. Different from the previous studies, this study proposes novel methods which does not require any image segmentation to perform simultaneously key-frame identification and LAV estimation. For this purpose, four different scenarios are designed, evaluated and compared using a 5-fold cross-validation strategy. Our collected dataset includes the apical four-chamber (A4C) view of 2-D echocardiography videos (621 videos with the resolution of 768 × 1024 pixels) taken from patients in Tehran Heart Center. The key-frames and their corresponding minimal and maximal LAV are determined by experts for this dataset. Experimental results show that the fourth proposed scenario outperforms the compared methods with an average accuracy of 93.68 ± 0.74 for key-frame identification and mean square error of 0.08 ± 0.01 for LAV estimation. The proposed scenario can be used in a fully automated manner for CAD software and mobile applications.

【 授权许可】

Unknown