Work

The Algorithms for Mapping Elastic Modulus (aMEM) Challenge

Year: 2026

Type: article

Abstract: Elastographic imaging is a powerful imaging technique that offers non-invasive estimation of the biomechanical properties of tissues. Ultrasound elastography methods have developed over the years, and this continues to be an active area of research in both clinical and preclinical domains where the main aims are to improve the detection and diagnosis of disease, but also to help understand the underlying mechanobiological mechanisms that promote disease. The Algorithms for Mapping Elastic Modulus (aMEM) elastography Challenge aimed to challenge the research community to process shear wave data obtained using ultrasound imaging to produce the best possible elastograms (maps of shear wave speed). Experimental imaging data obtained in phantoms, details on how the data were obtained and evaluation criteria was made available to research groups. The Challenge was organised in two categories: (i) acoustic radiation force impulse generated by the imaging probe, and (ii) harmonic vibrations generated by an external shaker. Participants were asked to submit elastograms, which were quantitatively analysed for background homogeneity (standard deviation in shear wave speed values), contrast-to-noise ratio, and accuracy in defining inclusion boundaries and location, for one soft and one hard inclusion (relative to background). The Challenge attracted entries from 16 research groups, particularly for the radiation force method, where a wide variety of methods were submitted, including deep learning approaches. The range of values obtained for homogeneity (0.05–0.76 for ARF, 0.08–0.9 m/s for harmonic), CNR (0.16–4.45, 0.3–4.28), inclusion edge resolution (2.65–9.14, 3.61–11.35 mm) and precision (0.26–5.54, 0.22–3.23 mm) evaluation metrics were found to vary greatly when excluding outliers. No individual method was found to stand out as clearly superior, with the best performing algorithms representing a variety of methods that included deep learning approaches. Results of the Challenge, including presentation of winners for each category, were presented at the workshop held at the joint IEEE-IUS Conference in Taipei, Taiwan. In addition to discussion of the various results, here, we also discuss fundamental questions and practical issues around running a challenge of this type.

Source: IEEE Open Journal of Ultrasonics Ferroelectrics and Frequency Control

DOI: 10.1109/ojuffc.2026.3724826