Executive Summary : | Ultrasound imaging, often referred to as "stethoscope of the future", is one of the most effective modalities for imaging of heart with advantages such as convenient, no ionizing radiation, real-time, and low-cost. Despite all these benefits, producing high-quality ultrasound images is difficult, particularly in the cardio space. This is because, depending upon the portion of the body selected for imaging, the sonographer must know where to place the ultrasound probe, how to orient it spatially, and finally, where to move it to acquire the desired image. Hence, the acquisition is dependent on the sonographer's skill level, thereby making it observer-dependent. Furthermore, the scarcity of skilled sonographer forces the available ones for repetitive work and imposes a heavy physical and cognitive burden, which not only affects image quality but also exposes the sonographers to health risks such as regional pain and musculoskeletal disorder. In addition to that, the frontline sonographers are susceptible to infectious diseases due to direct patient contact, particularly during pandemics such as COVID-19. Main objective of this project is to provide a solution to address all the aforementioned challenges. We will develop an echocardiographic image acquisition system for diagnostic and interventional use in which an ultrasound probe will be positioned (translational and rotational motion) by the joint coordination between an Artificial Intelligence (AI) model and a low-cost robotic system. More specifically, it is an AI-driven robot-assisted system for objective acquisition of clinically useful echocardiographic views. The system comprises a low-cost and portable robotic platform to hold the ultrasound probe, and a computer system to process the ultrasound images and run the AI algorithm to provide real-time navigation feedback to the robotic system. The AI model is primarily designed using deep convolutional neural networks to analyse the given image and evaluate whether the current image is correct in terms of quality and anatomical position. If the current image is not up to the mark, the model will provide feedback to the robot to adjust probe position and get the correct image. The given feedback will be executed by the robot having all six degrees of freedom (three rotational and three translational). The process ends when the desired quality image is acquired for a particular echocardiographic view. The project will involve development of an AI-driven objective echocardiographic view acquisition system, which will include creation of an echocardiographic standard views dataset as an intermediate step. This dataset will not only help in the development of the proposed system but also boost future research in the field of AI-driven echocardiography. |