The goal of this project is to develop a technique to transfer raw ultrasound data into either intermediate results or meaningful outcome (i.e., cardiovascular parameters) using neuromorphic computing.
What you will do
Ultrasound can be used to characterize cardiovascular function non-invasively. Recent advances in technology, i.e. the development of small ultrasound transducers (CMUT and PMUT), allows integration of ultrasound technology in patch-like form factor. Ideally, such a wearable ultrasound system should include real-time processing of raw ultrasound data. One option is to explore novel neuromorphic computing devices that offer lower energy consumption, and fast execution speed, while preserving acceptable accuracy when compared to standard computing systems (GPUs, CPUs, and microcontrollers).
The goal of this project is to develop a technique to transfer raw ultrasound data into either intermediate results or meaningful outcome (i.e., cardiovascular parameters) using neuromorphic computing. The goal is to reduce the data transfer rate from the cardiovascular patch without losing important features for the cardiovascular use case (i.e real-time monitoring of the blood pressure measured with ultrasound). To achieve this goal, the student will perform a literature research. The student will develop a deep neural network model for end-to-end ultrasound signal processing. The aim is to develop and benchmark spiking neural network architectures (recurrent, fully connected, and convolutional models), while exploring the tradeoffs of memory, accuracy, and computing requirements. After these developments, the student can map the selected spiking neural network in a prototype neuromorphic chip available in IMEC NL  and or in digital Field Programmable Gate Arrays (FPGA) boards. The student could also use a python framework already available at IMEC.
- Literature review (background ultrasound, cardiovascular function parameter and neural networks).
- Explore neural networks while exploring the tradeoffs for memory, accuracy, and computing requirements for the specific use case (ultrasound signal processing for medical applications).
- Implementation and training of spiking neural network.
- Testing developed neural network and analysis of results (e.g., benchmark in neuromorphic devices or in digital logic – DSP approach).
- Report and documentation.
What we do for you
You will be working on cutting-edge research on a topic that is relevant to both academic and industrial research groups. To help you in this journey, we offer a flexible environment where you can be the leader of your own research while at the same time have support of experts to complete your tasks.
IMEC has in-house experts in neuromorphic computing and ultrasound application who can help you in shaping this multi-disciplinary research project. In addition, there is an existing dataset that contains raw ultrasound, intermediate results and cardiovascular parameters ready to use, as well as, availability of neuromorphic device prototypes, and Field Programmable Gate Arrays for real-time execution of the spiking neural network model.
Who you are
- You are a MSc student in Electrical engineering, Data science of computer science.
- You are available for a period of 9 months.
- You have knowledge of neural networks through course and/or project work.
- You are excited about neuromorphic computing.
- You are entitled to do an internship in the Netherlands.
- You are self-starter and able to work independently.
- Good written and verbal English skills.
Click on 'apply' to submit your application. You will then be redirected to e-recruiting.