+1 on Reptilezs' answer.
Technically, you would need to measure the lux output at 1 meter over all 4 pi stearadian of solid angle around the flashlight and integrate it.
If you assume radial symmetry of the light beam, you could just measure the lux at all elevation angles relative to the light axis (with the axis of the beam being 90 degrees, the plane normal to the axis being 0 degrees, and the back of the light being - 90 degrees) and sum up that "slice" of measurements in very small increments with each measurement scaled by the cosine of the elevation angle as defined above.
Then you'd have to normalize the result against a similar measurement taken with a calibrated light source to convert your answer into lumens.
Most people who seek lumens estimates have some way of mixing all the light coming out of the box (central beam and spill) into a single light meter measurement which they then compare with the measurement of some known light to estimate lumens output based relative scaling.
The usual source for sort of standardized comparative measurements of flashlight output is the
FLR website, which includes directions on how the
lightbox used there is made. FLR took care of all this work for years; but sadly, the author Quickbeam has retired from this so no more new reviews will be made.