Keywords – Electrical impedance tomography; Microwave imaging; Stroke imaging; Inverse problems; Uncertainty quantification; Denoising / diffusion priors.
Context – The project framework is the diagnosis and follow-up of brain strokes that combines Electrical Impedance Tomography (EIT) [1] and Microwave Imaging (MW]) [2] so as the latter provides pertinent structural information to the former, guiding it to a more reliable result. In brief, via an efficient interplay of EIT and MWI, proper localization and classification under realistic acquisition variability is to be achieved. Acute stroke requires repeatable bedside monitoring to track disease evolution and treatment response. EIT is well suited to bedside use because it is safe, low-cost, and potentially fast; however, the reconstruction is challenging due to the severely ill-posed inverse problem faced with and strong sensitivity to practical factors like electrode placement and electrode–skin contact among others. MWI can provide complementary structural cues by exploiting dielectric/conductivity contrasts, although reliable reconstruction remains demanding in practice. Method development will start from public datasets and reproducible simulations, and will be consolidated through controlled phantom validation.
Project description – Acute stroke is a time-critical condition where clinical decisions depend on rapidly determining whether the event is hemorrhagic or ischemic, and—beyond the initial diagnosis—on monitoring how the patient evolves at the bedside. Current gold-standard imaging (CT/MRI) provides strong anatomical information, but it is not designed for portable, radiation-free, repeatable bedside monitoring. Two non-ionizing modalities are promising candidates to help fill this gap: EIT and MWI. This PhD, to summarize, proposes a multimodal, multiphysics reconstruction framework that leverages the practicality of EIT while using MWI-derived structural information to improve interpretability and robustness, with explicit uncertainty quantification to support reliable decision-making. EIT is safe, low-cost, and potentially real-time, making it convenient for bedside use. In this project, we will implement absolute/semi-absolute EIT reconstruction based on the Complete Electrode Model (CEM) [3], including explicit modeling of electrode contact impedances, and we will extend to multi-frequency EIT when data are available. However, brain EIT is a so-called ill-posed inverse problem. As a consequence, reconstructions from collected data are highly sensitive to practical acquisition factors like electrode placement and contact impedance variability, and spatial resolution is inherently limited. These limitations suggest that EIT alone may not consistently provide sufficient structural detail for reliable lesion localization, especially under realistic electrode variability. A natural way to strengthen EIT is to incorporate complementary structural information from a second modality: MWI. MWI can provide electrical permittivity contrasts at higher frequencies and is better positioned to recover internal structural information.
The central methodological challenge is to combine these two modalities in a way that improves EIT while remaining faithful to EIT measurements. We will explore physics-guided fusion strategies in which learned priors are integrated into EIT reconstruction through iterative optimization. A primary route is Plug-and-Play (PnP) reconstruction [4], where a learned denoiser (Convolutional Neural Network (CNN)-based or diffusion-based) acts as an implicit within an iterative solver. Moreover, a Bayesian perspective [5] wherein the forward model defines the likelihood and the learned prior induces a structured regularization will be explored.
Beyond reconstruction quality itself, the PhD will explicitly address a key requirement in safety-critical medical imaging: reliability [6],. This will be achieved by producing uncertainty or confidence indicators (pixel- or ROI-level) together with acquisition quality metrics that can flag unstable reconstructions, detect electrode-related issues, and guide interpretation.
Overall, the PhD will deliver an EIT-centered multimodal reconstruction framework that is structure-guided, robust to electrode effects, and uncertainty-aware, leveraging complementary information from MWI. Validation on public data [7], simulations, and a dedicated phantom platform should establish a clear pathway toward future bedside translation.
Supervision – Thomas Rodet, Full Professor, SATIE, ENS Paris-Saclay, thomas.rodet@ens-paris-saclay.fr; Yarui Zhang, Associate Professor, SATIE, ENS Paris-Saclay, yarui.zhang@ens-paris-saclay.fr.
Candidate profile and required skills – Master’s degree or diplôme d’ingénieur in one or more of the following areas: signal and image processing, applied physics, applied mathematics and computational modeling. Experience in deep learning and its ecosystems (libraries/tools/languages). Prior exposure to medical imaging welcome.
References :
[1] Yang, L., Xu, C., Dai, M., Fu, F., Shi, X., & Dong, X, « A novel multi-frequency electrical impedance tomography spectral imaging algorithm for early stroke detection, » in Physiological Measurement, 37(12), 2317, Dec. 2016.
[2] R. Scapaticci, J. Tobon, G. Bellizzi, F. Vipiana and L. Crocco, « Design and Numerical Characterization of a Low-Complexity Microwave Device for Brain Stroke Monitoring, » in IEEE Transactions on Antennas and Propagation, vol. 66, no. 12, pp. 7328-7338, Dec. 2018.
[3] Vauhkonen, P. J., Vauhkonen, M., Savolainen, T., & Kaipio, J. P., « Three-dimensional electrical impedance tomography based on the complete electrode model, » in IEEE Transactions on Biomedical Engineering, 46(9), 1150-1160, 2002.
[4] U. S. Kamilov, H. Mansour and B. Wohlberg, « A Plug-and-Play Priors Approach for Solving Nonlinear Imaging Inverse Problems, » in IEEE Signal Processing Letters, vol. 24, no. 12, pp. 1872-1876, Dec. 2017.
[5] Holden, M., Pereyra, M., & Zygalakis, K. C, « Bayesian imaging with data-driven priors encoded by neural networks, » in SIAM Journal on Imaging Sciences, 15(2), 892-924, 2022.
[6] Swaminathan, M., Bhatti, O. W., Guo, Y., Huang, E., & Akinwande, O. , « Bayesian learning for uncertainty quantification, optimization, and inverse design, » in IEEE Transactions on Microwave Theory and Techniques, 70(11), 4620-4634, 2022.
