Next gen ultrasound imaging for rehabilitation robots

Today, state-of-the-art robots use human-machine interfaces that sense muscle activity within the body by reading the electrical nerve signals at the surface of the skin. Although this approach offers a fast readout, it suffers from two major drawbacks: 1. The surface electromyography or sEMG sensors have a low signal to noise ratio since they detect electrical signals generated at the nerve-muscle junctions located deep below the skin and 2. The electrical nerve spiking signal of interest is not intuitive when used for actuation purposes like moving a robotic arm. 

Ultrasound imaging can offer a novel solution to this problem since it can look at the actual deflection of muscle fibers within body tissues. In our MMICRO group, we achieve this imaging using special miniaturized ultrasound sensors called pMUTs – piezo electric micromachined ultrasound transducers. The novelty of these pMUT devices is that they can work as both transmitters as well as receivers of ultrasound waves by converting alternating electrical signal to mechanical vibrations of a membrane and vice versa i.e., the piezoelectric effect. The advantages of using these pMUTs for imaging is that that they offer high-resolution ultrasound imaging while maintaining a small form factor that allows better integration with existing technologies. 

Figures: (left) Forearm under ultrasound [adapted from N. Akhlaghi et al, IEEE Trans on Biomed Engg], (right) Working principle - pMUT device [youtube link] 

At imec, we are working on a project to integrate these pMUTs with the advances in machine learning approaches to sensor data analysis. The idea is to combine the complexity of ultrasound signal data with appropriate target classification algorithms to achieve a machine-learning-aided pMUT sensor. We are currently working on a proof-of-concept mm-sized bulk ultrasound sensor using machine-learning approaches. And this is where the role of the master’s student comes in.  

The goal of this master thesis proposal is to benchmark the performance of our machine-learning aided bulk transducers to existing commercial electrical sensors. This project will give an opportunity to master the basics of ultrasound imaging and understand advanced machine learning algorithms used for target classification. This would also involve working with commercial electrical sensors and research ultrasound systems. High creativity and interdisciplinary approach to working is much appreciated. The work is estimated to be: 70% characterization, 30% data analysis.