Mathematically the ray tracing and image source methods are the same, except in one instance. If the image sources are such that a series can be established that has a known answer then this is equivalent to the modal summation method. This can really only occur in rectangular rooms or some cylindrical ones.

But, if the images are not so arranged, i.e. the room is not rectangular, then the image method will converge very slowly at low frequencies and it would take a very large number of images to get an accurate answer.

I would suggest a test case where I use the modal method, which we would have to agree is the accepted standard answer, and you can run CARA. I would be most interested in this comparison.