We don't have that function exactly, but it's fairly easy to do what you want a few ways:
# use the atlas dataset directly
3dcalc -a ~/abin/MNIa_caez_ml_18+tlrc.'<92..116>' -expr a -prefix atlasextract
# use a concatenated master dataset as a single dataset
3dcalc -a "CA_ML_18_MNIA::91 CA_ML_18_MNIA::92 CA_ML_18_MNIA::116" -prefix threeregions -expr a
# concatenate datasets with 3dbucket or 3dTcat
3dbucket -prefix buck_atlas "CA_ML_18_MNIA::91" "CA_ML_18_MNIA::92" "CA_ML_18_MNIA::116"
# combine with 3dcalc
3dcalc -a "CA_ML_18_MNIA::91" -b "CA_ML_18_MNIA::92" -c "CA_ML_18_MNIA::116" -prefix threeregions_1sb -expr "step ( a ) +2*step ( b ) +4*step ( c ) "
The first method keeps the original values of 92 to 116 in a single sub-brick. The next two give multiple sub-bricks that have values of 1 for each region and sub-brick labels are the original values. The last method creates a single sub-brick dataset and uses a binary assignment using values of 1,2 and 4 for individual values (other values would show overlap which you wouldn't get for this particular atlas).