The output dataset contains a 32-bit-real-valued image which is that linear combination of the input model components which most closely approximates the expectation value of the (masked) image . The best-fit amplitudes and the names of the component datasets are stored in the image array header in a table named COMPDATA which has columns as follows: