I suspect your scanner or sequence was upgraded. Classically, reconstruction was 12 bits (0..+4095), but recent upgrades have moved to 16 and 20 bit precision. My software always tries to uses the most compact popular format possible. So if your raw data was in the range -32768..+32768 it will be stored as signed 16-bit integers. If your data is in the range 0..65535 it will be saved as either unsigned 16-bit integers or 32-bit floats (depending on your ini files setting "UINT16toFLOAT32=0", "UINT16toFLOAT32=1": the former is more compact and the latter is better supported). If your data range exceeds 65535 the image will be stored as 32-bit float.

You can always upsample 16-bit data to 32-bits (though they take twice as much disk space). For example, my script
 https://github.com/rordenlab/spmScripts/blob/master/nii_change_datatype.m

I would also suggest upgrading to dcm2niix (which comes with MRIcroGL). While I hope my older software is robust and mature, I have moved my development efforts to the newer software. The scanner manufacturers are re-interpretting the DICOM standard, so be vigilant that newer updates may break my tools.