Robotic Standard View Acquisition in Transthoracic Echocardiography Using Skill Imitation Learning Method

Transthoracic echocardiography (TTE) is the most common and convenient screening method for diagnosing heart disease. In freehand echocardiography, the physician guides an ultrasound (US) probe to visualize all standard views. However, echocardiography is the most challenging task in ultrasound examination because of the complicated acoustic windows and anatomical structures, which depends on extensive clinical experience of the physicians. To address this challenge, this study proposed a robotic system for semi-autonomous echocardiography acquisition. Once the physician remotely guides the probe to the position near a target acoustic window, the robot then fine-tunes the probe autonomously to obtain the standard view. To achieve autonomous intelligence of robotic scanning, we presented an imitation learning based deep neural network to model the ultrasound physicians' scanning skills. To evaluate the effectiveness of the proposed method, operation data were recorded during the physician's scanning demonstration to train the policy of auto-navigation. Taking the parasternal long-axis view as an example, preliminary experiments on a heart phantom validated that the proposed approach can effectively and accurately navigate the probe towards the standard views of TTE.