Noise in SPECT is usually a count problem wearing a different label. The order to work through: administered activity and acquisition, collimator and energy window, uniformity and centre of rotation, reconstruction settings, then the patient and the room.
Noise in a SPECT study is nearly always a count problem wearing a different label. Something reduced the counts, something spread them out, or something in reconstruction amplified what statistical variation was already there. Work through it in the order below, change one thing at a time, and write down what you changed. A department that adjusts three settings at once has not solved the problem, it has moved it somewhere harder to find.
Confirm which collimator is physically fitted, on which head. A high resolution collimator on a low-count study is a legitimate and frequent cause of a noisy result, and swapping to a general purpose collimator is sometimes the whole fix. On a dual head, check both, because a mismatched pair produces an oddly inconsistent image that reads as noise.
Now the energy window. Check that the peak has not drifted, that the window width is what the protocol specifies, and, most importantly, that the isotope selected in the protocol is the isotope in the patient. A window set for one photopeak and an acquisition of another is a spectacular way to lose counts, and it happens on the day a department runs an unfamiliar study.
Look at today's flood. Then look at the trend over the last three months, because a slow drift is invisible in a single image. Ring artefacts in the reconstruction are the classic sign of a uniformity problem, and they are routinely reported as noise by people who have not seen them before. Check when the high-count uniformity correction map was last acquired, because an out-of-date correction is worse than none.
Then centre of rotation and head alignment. A drifted centre of rotation blurs the reconstruction in a way that looks like poor statistics. On a dual head, misalignment between the two detectors produces the same appearance. Both are measured, not judged by eye, and both should have a record showing when they were last verified.
A failing photomultiplier tube will usually show on the flood before it shows anywhere else, which is the argument for actually looking at the daily flood rather than acquiring it and filing it.
This is the commonest software cause and the easiest to test, because you can reprocess the same raw data repeatedly.
Review the cine and the sinogram before the patient leaves the table. Motion is visible there and nowhere else once the study is reconstructed, and a patient who moved is a patient you can sometimes still re-acquire. Look for arms in the field, metal, and positioning that put the region of interest at the edge.
And the room itself, which matters more in this climate than the textbooks suggest. Detector electronics drift when they run warm, and an air conditioning unit that failed overnight will show up as a system that peaks differently in the morning. Humidity and unstable mains are the two environmental causes worth ruling in early, because they explain a pattern of problems rather than a single bad study.
When you think you have found it, prove it on a phantom rather than on the next patient. Re-acquire the same phantom at the same settings you had before the change, compare against the baseline you filed at commissioning, and record the result in the QA book. That baseline is the reason the department can answer this question in an afternoon instead of a fortnight.