Dear Colleagues,

We are offering an M2 internship titled "Super-Resolution in Ultrasound Imaging Using Diffusion Models" at the Institut de Recherche en Informatique de Toulouse (IRIT) in Toulouse.

--------------------------------------------------------------------------------------------------------------------------------

Ultrasound (US) imaging has become increasingly important in medical practice due to its non-invasive nature, lack of ionizing radiation, cost-effectiveness, and ease of use. However, US wave propagation is subject to physical limitations, such as diffraction and attenuation, which constrain spatial resolution. To address this limitation, various approaches have been proposed, including US image super-resolution (SR) techniques, which typically convert low-resolution (LR) images to high-resolution (HR) outputs [1], [2]. Nevertheless, these methods often rely on mathematical models that require empirically tuned hyperparameters. Recently, deep learning (DL) solutions have shown promise in enhancing SR performance, offering improved imaging efficiency and faster computation. The strength of DL-based methods lies in their ability to learn parameters directly from data, allowing for accurate and rapid SR imaging. However, a significant challenge in applying DL to US imaging is the limited availability of training data, especially the lack of ground truth labels (e.g., data labeled by experts) for in vivo data. To address this, some studies use outputs from optimization algorithms as ground truth after fine-tuning or use data from high-performance US systeme. Additionally, part of the training is often conducted on simulated data, with a growing trend toward using highly realistic simulations based on optical images of microvasculature and in vivo data.

In this internship, the focus will be on enhancing US imaging resolution through SR techniques that combine an unsupervised approach, known as Denoising Diffusion Restoration Models (DDRM), with existing US SR work [1], [2]. DDRM leverages a pre-trained denoising diffusion generative model to address a range of linear inverse problems, offering a promising pathway to improved imaging resolution [3].

References:
[1] N. Zhao et al., “Fast single image super-resolution using a new analytical solution for l2-l2 problems,” IEEE TIM, vol. 25, no. 8, pp. 3683–3697, Aug. 2016. doi: 10.1109/tip.2016.2567075.
[2] N. Zhao et al., “Single image super-resolution of medical ultrasound images using a fast algorithm,” in 2016 IEEE ISBI, 2016, pp. 473–476. doi: 10.1109/ISBI.2016.7493310.
[3] B. Kawar et al., Denoising diffusion restoration models, 2022. doi: 10.48550/ARXIV.2201.11793.
--------------------------------------------------------------------------------------------------------------------------------

For more information, please visit: https://cloud.irit.fr/s/itHhl0aqwg9DcJB.

Application: Prospective candidates should submit the following documents as a SINGLE PDF file:

1. University transcripts
2. A one-page cover letter
3. Curriculum vitae (CV), including any publications if applicable
4. References

Please email the application to duong-hung.pham@irit.fr and denis.kouame@irit.fr.

Feel free to share this opportunity with any interested students.

Thank you very much.

Best regards,

--

PHAM Duong Hung, Ph.D

UNIVERSITÉ DE TOULOUSE

IRIT Lab, UMR 5505 – Office 209

118 Route de Narbonne, 31400 Toulouse, France

Phone: (+33)0 5 61 55 65 99

Mobile : (+33)0 7 68 49 68 85

Url: irit.fr/~Duong-Hung.Pham