Recent advancements in helmet-type magneto-encephalography (MEG) systems that operate without liquid helium, such as optically pumped magnetometer (OPM)-based MEG, have increased interest in MEG-based brain-computer interfaces (BCIs). Among various BCI paradigms, steady-state visual evoked field (SSVEF)-based BCIs have been actively studied owing to their high information transfer rate (ITR) and low demand for calibration sessions. Although MEG provides excellent spatial resolution and whole-head coverage, conventional algorithms such as the filter bank-driven multivariate synchronization index (FBMSI) do not fully exploit these advantages. To overcome this limitation, this study employed MEG source imaging to utilize information from whole-head MEG recordings fully and developed a novel weighting method called the averaged source location-based weighting (ASLW). ASLW leverages the averaged source locations of SSVEF signals to enhance BCI performance. Experimental results with 20 participants demonstrated that integrating ASLW with the FBMSI algorithm (ASLW-FBMSI) significantly improved both the classification accuracy and ITR across all tested window sizes. Notably, the largest performance gains included a 13.9% accuracy improvement at a 3-s window size and a 13.1 bits/min increase in ITR at a 2.5-s window size. Additionally, the ASLW-FBMSI algorithm exhibited a short processing delay of 0.107 s at a 4-s data length and was successfully validated in online BCI experiments with 20 participants. Although tested in SQUID-MEG in this study, our findings demonstrate the effectiveness of ASLW in significantly enhancing the overall performance of SSVEF-based BCIs.