In principle, Huygens Deconvolution MLE algorithms keeps the integrated intensity of the image. The process can be interpreted as the algorithms re-localizing the light distributed in the 3D volume, gathering the blur and bringing it back to its original source. Like that, the total intensity (integrated all over the volume) in the restored image has an expected value equal to the original's. This is due to the Point Spread Function having a total integrated intensity of 1.
Still, the PSF's normalized intensity is not the only factor responsible of the final result. Things that may alter the integrated intensity in practice are
1) A BackGround removal. The background signal as set by the parameters is gone after deconvolution, and can't contribute to the total intensity anymore. It will look as if it was reduced. Therefore the correct sentence would be "integrated intensity of the interesting signal is preserved". If you want to preserve all the intensity, set the background to zero in the Restoration Parameters.
2) A constant multiplicative factor (reported by the software) if the memory of the computer is limited and the data must be kept at 16 bit. This factor is usually smaller than 1 if the intensity of the restored maxima increases so much during the deconvolution that the actual values exceed the 16 bit range. To keep it within that range, a normalization factor may be applied that reduces the integrated intensity. See Ratiometric Images. When there is sufficient memory available and the system is 64 bit, the final result is in 32 bit floating point format which requires no further scaling.
3) Blur that is too close to the image borders, both in the lateral as in the axial directions, may be restored as originating from sources outside the image. These sources won't be in the restored result, not contributing to the integrated intensity. Thus, light can "leak out" at the image limits, if the imaged objects are "beyond" them!!! This effect is usually very small if the imaged volume is large in all dimensions and your objects are away from the borders, but may become appreciable in for example flat Wide Field Microscope and shallow confocal images.
4) As a consequence of the previous point, images that for some reason must be deconvolved in independent blocks (Brick Splitting) will suffer that effect multiple times, once per block: the total amount of light that leaks out is higher, for the benefit of a higher blur removal.
5) Storing the result in some file formats with Dynamic Range limitations. See Tiff Scaling.
6) 'Defective' clipped raw data that does not represent physically realistic intensities.