1st supervisor: Kawal Rhode, King’s College London
2nd supervisor: Christos Bergeles, King’s College London
Cardiovascular disease (CVD) has a significant impact on society in terms of mortality, morbidity and healthcare costs. For example, it causes over 1.9 million deaths in the EU every year (42% of all deaths) with a total estimated cost of £118 billion. Medical imaging has a very important role to play in the diagnosis, treatment guidance and follow-up of patients with CVD. Cardiac ultrasound or echocardiography allows real-time visualisation of target heart structures and interventional devices without using ionising radiation. This can be done using a TTE or TOE and is often the first-line imaging modality for CVD. The training and expertise required to perform echocardiography is considerable. This, coupled with the high demand for this type of imaging, means that healthcare providers are becoming increasingly strained to provide this service despite the comparative low cost of the hardware. We have developed a range of robotic ultrasound systems including single- and multiple-arm TTE and TOE. These could have significant impact if they could operate in a highly-autonomous manner with minimal operator input. Artificial intelligence algorithms for dealing with complex inverse kinematics, recognizing standard ultrasound views of organs and analyzing medical image quality have emerged over the recent years. These algorithms could be combined in a closed loop system to develop an intelligent robotic echocardiography system. The challenges are to effectively combine these elements and to validate and test from bench to bedside.
Months 1-3: Literature review and ethics
Review of literature on robotic ultrasound systems, artificial intelligence to solve large DOFs serial chain and flexible robotic inverse kinematics and artificial intelligence to analyze ultrasound image quality. Apply for KCL Research Ethics for a healthy volunteer study on robotic echocardiography.
Months 4‐9: AI-driven view classification
Evaluate a CNN for identification of standard 2D cardiac echo views using retrospective data from adult cardiology and/or healthy volunteer data (TTE only for latter). This will be limited to all standard TTE views and approximately 5 TOE views (out of the possible 20). Submit paper 1.
Months 10-15: Open loop robotic steering
Develop and validate AI-based inverse kinematics solutions for our existing ultrasound robots. Validation will be carried out using a target object phantom and EM tracking to determine the accuracy and robustness of the steering.
Months 16-21: Closed loop robotic steering
Couple and validate AI-based view classification with AI-based inverse kinematics. Validate this using an anthropomorphic cardiac phantom. Test this approach using healthy volunteers and single-arm TTE robotic system. Submit paper 2.
Months 22-30: Fine control
Evaluate methods for measuring echo image quality and incorporate into the closed-loop robotic steering. Validate this using an anthropomorphic cardiac phantom. Test this approach using healthy volunteers and single-arm TTE robotic system.
Months 31-36: Extensions & write-up
Possible adaptation to multi-arm or concurrent TTE and TOE systems. Write up thesis. Submit final paper.
