arXiv:2611.22267v1 Announce Type: new Abstract: MRI is not designed to operate in the presence of EMI emitting electronic devices. To accomodate this, we propose VERGE which improves on previous EMI removal methods and can operate without external EMI sensors. VERGE estimates a reference coil non-linearly from all imaging and non-imaging coils by training a small per-scan convolutional neural net to generate a virtual coil. The virtual coil is then used with STRIDE, a reference-based EMI removal method. For single-source EMI contamination, VERGE performed similarly to the reference method. For multi-source dynamic EMI contamination, VERGE outperformed the comparison methods. Notably, VERGE with zero physical EMI sensors outperformed the comparison methods even with two physical EMI sensors. A non-linearly estimated virtual EMI coil improves reference-coil-based EMI removal under dynamic multi-source EMI, and enables effective EMI removal without physical EMI sensors or training data.