Stefano Mandija, Alessandro Arduino, Cornelis A. T. van den Berg, Patrick Fuchs, Ilias Giannakopoulos, Yusuf Ziya Ider, Kyu-Jin Jung, Ulrich Katscher, Dong-Hyun Kim, Riccardo Lattanzi, Thierry G. Meerbothe, Khin Khin Tha, Luca Zilberti
4 min
Magnetic Resonance Electrical Properties Tomography (MR-EPT) is a technique that aims to map the electrical conductivity of biological tissues using standard MRI hardware. Despite its potential for clinical diagnostics, the field has lacked a standardized way to evaluate the various mathematical algorithms used to reconstruct these maps. This study presents the results of "Phase 3" of the first MR-EPT reconstruction challenge, which sought to benchmark these methods using real-world, measured data rather than idealized simulations.
The researchers provided participants with a standardized dataset consisting of MR measurements from three distinct sources: a simple cylindrical phantom, a complex 3D-printed brain phantom, and a healthy human volunteer. To ensure a robust test, the transceive phase—a critical component for calculating conductivity—was captured using three different MRI acquisition sequences: bFFE, Spin-Echo, and Turbo Spin-Echo. A total of 38 participants submitted their own reconstruction algorithms to process this data, allowing for a comparative analysis of how different mathematical approaches handle real-world noise and measurement artifacts.
The challenge revealed a significant lack of consensus in the field. The submitted conductivity maps exhibited high variability in both visual quality and quantitative accuracy. Expert reviewers found that performance—measured by precision and the ability to accurately reconstruct tissue boundaries—differed drastically between participants. Notably, many algorithms struggled to maintain consistency across the different acquisition methods, suggesting that the choice of reconstruction technique is highly sensitive to the underlying data source.
This study highlights a critical "bottleneck" in the development of MR-EPT: the absence of a gold-standard reconstruction method. The large variations observed among the 38 submissions demonstrate that current techniques are not yet interchangeable or universally reliable. By establishing this benchmark, the authors provide a necessary framework for the research community to identify which algorithms are most robust. This is a vital step toward moving MR-EPT from an experimental technique toward a reliable, standardized tool for clinical imaging and non-invasive tissue characterization.
Motivation: To benchmark MR-Electrical Properties Tomography (MR-EPT) reconstruction methods. Goal(s): To present the results of phase 3 of the MR-EPT reconstruction challenge. Approach: MR measurements were performed for a cylindrical phantom, a 3D printed brain phantom, and a healthy volunteer. In each case, the transceive phase was measured using three different methods: bFFE, Spin-Echo, and Turbo Spin-Echo acquisitions. Results: 38 participants submitted conductivity reconstructions. Not all participants reconstructed conductivity maps for all three different transceive phase maps. The reconstructed conductivity maps show large variability in quality (visual scoring by experts), and performance (accuracy and precision). Impact: The results of phase 3 of the first MR-EPT reconstruction challenge show large variations in the estimated conductivity maps from measured data on phantoms and volunteer demonstrating the need of benchmarking reconstruction methods on common datasets.
Sam: Not even close. Despite starting from identical data, the 38 teams produced maps with large discrepancies. Some were sharp and accurate. Others were blurry or showed significant errors — particularly at the boundaries where one type of tissue meets another. Those boundary regions are often exactly where you'd most want precision, since that's where tumors or abnormalities tend to appear.
Alex: So the bottleneck isn't the MRI machine itself — it's the mathematical method each team uses to interpret the signal. Like having the same photograph but running it through 38 different filters, each one producing a different image.
Sam: That's a precise way to put it. Each team uses their own algorithm — their own set of mathematical rules — to translate the wave-shift data into a conductivity map. And because those algorithms differ, so do the maps. The study suggests that until the field agrees on a standardized approach, this kind of electrical mapping remains difficult to trust in a clinical setting. We have the data. We just don't yet have a single, reliable way to read it.
Alex: Which means the next step isn't better hardware — it's better agreement on the software side.
Sam: Precisely. That's what this challenge was designed to push toward. By showing just how wide the variation is across expert teams, the researchers are making the case that standardization needs to happen before electrical property mapping can move from research labs into hospitals. The data is there. The question is whether the field can converge on a shared method for interpreting it.
Alex: Thanks for listening to ResearchPod.