Higher accuracy was achieved in an earlier study where another team used large fMRI machines (it was featured in the version for May). There participants listened to audiobooks / speech while being in the large machine; I guess long training would be easier here but it’s more limited since it’s EEG. However, they claim they have exceeded 60% by now.
If you don’t understand the study itself or in general if you’re interested in it, it’s always a good idea to also read a good news report on it; see this and also this. They found carrot intake rather than beta-carotene, the focus of prior studies, has this association and figure 6 was just to show that they don’t have much data on daily intake of a carrot or more.