Electrical impedance tomography (EIT) is an emerging technology for real-time monitoring of patients under mechanical ventilation. EIT has the potential to offer continuous medical monitoring while being noninvasive, radiation free, and low cost. Due to their ill-posedness, image reconstruction typically uses regularization, which implies a hyperparameter controlling the tradeoff between noise rejection and resolution or other accuracies. In order to compare reconstruction algorithms, it is common to choose hyperparameter values such that the reconstructed images have equal noise performance (NP), i.e., the amount of measurement noise reflected in the images. For EIT many methods have been suggested, but none work well when the data originate from different measurement setups, such as for different electrode positions or measurement patterns. To address this issue, we propose a new NP metric based on the average signal-to-noise ratio in the image domain. The approach is validated for EIT using simulation experiments on a human thorax model and measurements on a resistor phantom. Results show that the approach is robust to the measurement configuration (i.e., number and position of electrodes, skip pattern) and the reconstruction algorithm used. We propose this novel approach as a way to select optimized measurement configurations and algorithms.