So to wrap up this thread for the interested public -  it turns out that Inf/NaN values for SNR and CNR will arise when the dynamic range of the input images is truncated, such that all CNR and the SNR sphere have identical and constant intensity values (e.g., 4095 if 12bits are stored but the actual values pre-DICOM were significantly higher).

In that case, SNR is Inf because noise is computed as zero (all values are 4095), and the CNR NaN values arise from zero contrast divided by zero noise.

Torsten