The limitations of contact-based hardware for cardiac activity measurement in MRI (e.g. electrocardiography, pulse oximetry) might eventually be overcome by using videos of the skin. Currently, a limitation for practical usage is the poor signal-to-noise ratio of the acquired signals, especially in an MR environment. We analyzed pixel intensity variations over time measured on the forehead and observed that 15%-25% of pixels exhibit a different signal morphology and phase than the remaining pixels. This effect is a limitation for many algorithms related and unrelated to MRI: If all pixels within a region-of-interest are used for averaging, different signal types are mixed which biases results.