Image itself is in Spatial domain then why we need to go to frequency domain? because we want to do some processing in frequency domain (which were not possible or were difficult in spatial domain). so we take Fourier Transform of the image, do some processing in Fourier domain and bring back to Spatial Domain.
why we need to bring back? because we need the processed image, not the Fourier transform of the processed image.
Totally depends on your application. You can use the fourier transform to analyze or filter your image. In case of the former you need not transform it back to the spatial domain. In the later case, you may need to further process the spatial content of your image which will need you to carry out the inverse fourier transform.
On top of other answers, you could also create the image directly in frequency domain, as is the case of Magnetic Resonance Imaging (MRI). Then, naturally, you need to use the Inverse Fourier Transform to get the spatial image.