Research Internship - Development of a deep learning method for ultrafast localization microscopy

Objectives of the project:

To develop and validate a deep learning framework for precise localization and tracking of microbubbles in ultrafast ultrasound images, for Ultrafast Localization Microscopy

State of the art & rationale:

Ultrafast Localization Microscopy is a recent technological breakthrough allowing for super-resolution imaging of the vasculature using the precise localization of microbubbles injected into the blood flow [1]. The technique is currently limited by the need for long acquisition times or high microbubble concentration, mainly because of the low sensitivity of microbubble detection. Deep learning techniques have the potential to overcome current limitations of physics-based algorithm by generalizing better and providing faster results [2]. A deep learning model developed in our laboratory has already shown promise for precise localization of a single microbubble, and the aim of this project is to extend it to multiple microbubbles.

Methods and tools:

A convolutional neural network architecture will be developed using the Python language and simulated data already available for the project. The algorithm will be validated using the simulation as ground truth, and tested on in-vitro phantom experiments and ULM experiments on rodents.

Main tasks:

• Algorithm design and implementation
• Benchmarking against state-of-the-art classical localization tools
• In-vitro validation
• In-vivo validation

Anticipated outcomes and potentials:

The method developed will be published in a peer-review journal, and will be implemented in the ULM pipeline available at our laboratory.

Role of the student:

The student will be responsible for choosing and implementing the algorithm, and will be involved in designing and carrying out the experiments, as well as writing up the results.

Duration: 6 months

Location: Institute Physics for Medicine Paris, 2-10 rue d’Oradour-sur-Glane, 75015 Paris

Contact: beatrice.walker@espci.fr, justine.robin@espci.fr

Bibliography:

[1] C. Errico et al., « Ultrafast ultrasound localization microscopy for deep super-resolution vascular imaging », Nature, vol. 527, no 7579, p. 499‑502, nov. 2015, doi: 10.1038/nature16066.
[2] X. Chen, M. R. Lowerison, Z. Dong, A. Han, et P. Song, « Deep Learning-Based Microbubble Localization for Ultrasound Localization Microscopy », IEEE Trans. Ultrason., Ferroelect., Freq. Contr., vol. 69, no 4, p. 1312‑1325, avr. 2022, doi: 10.1109/TUFFC.2022.3152225.