There's no really good way, especially if the vocals have significant stereo-spreading fx (such as reverb or delay) active in a non mono-centered fashion, and/or if the vocals are not themselves panned dead-center.
In case the material has centered, dry enough vocals, then import back the mixdown into Logic on an empty audio track, insert a Gain plug-in, and enable the invert phase button on either the L or R (not both!) channel. This should more or less null everything that is centered (among which, the vocals), only leaving the L- or R-specific (i.e. non-centered) audio signal.
Doing so, you may well null more than the vocals (e.g. bass, kick, possibly snare if centered as well...). If that's the case you have to do this using parallel tracks along with bandwidth EQing, to for instance leave the bottom end of the signal (e.g. below 200 Hz, where the bass and kick sit) unprocessed, while processing only the bandwidth in which the vocals main frequencies lie.
This is most probably going to be less than perfect, or even close to impossible, depending on the material. I think that most decent karaoke tracks are actually remixes where the vocals were muted from within the mix itself.
Logic Pro X 10.4.8 - OS X Catalina 10.15.3
MacBook Pro retina 16" late-2019, 2x2.4 GHz Core i9, 64 GB 2666 MHz DDR4, 8 TB SSD
Audio interface: RME Fireface UCX - MIDI Interface: MIDIMAN USB Midisport 4x4
Keyboard midi controllers: Korg M1, NI Komplete Kontrol S88 mk1
Additional control surface: Logic Remote (on iPadAir 2)https://www.evernoize.com