Provide heavily_filtered VDD to the LNAs. That reduces one source of external noise.
And then use a copper or steel shield atop the two LNAs, to exclude electric and magnetic effects.
In general the direction of interfering fields will not be known, so simply orienting the two LNAs in opposite directions will not, in the general case, suffice.
However, I was told, as a much younger engineer, that magnetic fields could not be shielded.
That is wrong.
The induced voltages resulting from changing magnetic fields are simply the displacement currents that attempt to provide nulling charges in response to electric fields.
Thus all the concepts of skin effect become useful in attenuating "magnetic" fields.
About 30 years ago, needing to develop a data acquisition system with 15_bit dynamic range, with about 30MHz bandwidth in the opamps, I took an afternoon to evaluate how to attenuate high frequency wire_wire coupling in the form of 1" square loops.
And I used a commonly available source of aluminum foil: chewing gum wrappers.
The test waveform was 20 nanosecond Trise and Tfall square waves, perhaps at 1MHz. Driving a leaded 51_ohm resistor, leads in 1" square loop and soldered onto Coax cable.
I built two of these 1" loops, to be Transmitter and Receiver/
With 1 volt pulses driving the Transmitter loop (thus 1/51 = 20 milliAmps in 20 nanoSeconds), I measured 0.08 volts in the Receiver loop, those two loops held about 1/16" part.
And after wrapping the foil between a folded sheet of paper, to prevent any electrical contact, I slowly slide the "shield" between the TX and RX loops.
Result? about 30dB drop in amplitude, or about 3 milliVolts on the TEk7904 scope's 7A26 50Ohm Rin 400MHz bandwidth plugin.
Do your own experiments, but think about metallic shields, to greatly attenuate both electric and "magnetic" interference.