Simultaneous measurements by surface electromyography (sEMG) and diffusion-weighted magnetic resonance imaging (DW-MRI) enable studies of small muscular activities in resting human musculature. Induced signal disturbances can hamper the evaluation of small amplitude myoelectric signals despite the application of gradient artifact correction (GAC) methods. Systematic evaluation of the quality of GAC methods is difficult due to the unknown real sEMG signal during simultaneous measurements. Therefore, a new concept for GAC quality estimation is investigated based on sEMG sample generation and classification. Signal synthesis and classification is based on a Dense-Bidirectional-Long-Short-Term-Memory (DBLSTM)-based auxiliary classifier generative adversarial network (AC-GAN). Quality of GAC and ability to detect small myoelectric activities is exemplary investigated on two common DW-MRI sequences.