What is your technique for MRI to CT fusion for intracranial SRS planning?

Acurrate fusion has become a real concern with single isocenter VMAT and mulitple small peripheral mets, as there are many ROIs. Todd, the abstract you cited for fusion on the pediatric protocol: 1.8 mm average error with 2 mm standard deviation (https://www.ncbi.nlm.nih.gov/pubmed/20381270)

Presently, I use Eclipse for fusion which does not seem as capable as brainlab, mim etc for autofusion. To optimize fusion, we try to obtain a planning MRI with orthogonal slices (no gantry tilt) 256 x256 MPRAGE (volumteric study) on 1.5 T MRI (we were getting more spatial distortion on our 3T). For the CT simulation, I have found that IV contrast can really help with the fusion. The flicker function also is more helpful than the sliding window for detecting small differences in alignment given our natural tendency to follow lines.

For some of these cases, I am more comfortable with a 3-5 fraction schema as it allows 1-3 mm margins.

Rigid based registration is performed for cranial lesions while deformable registration may be required for extra-cranial sites.  Among other factors the agreement accuracy will be affected by the imaging site and technique, magnitude of rotational discrepancies between the studies, patient motion and the extend of the scans.  Therefore, one should never assume an automated registration will yield acceptable results and always confirm satisfactory agreement has been achieved.  Using bone and soft tissue windows/levels the alignment is verified by inspecting clearly delineated regions of interest in the axial, sagittal and coronal views, with emphasis on the treatment area.  If disagreement is beyond the desired tolerance, it is recommended to adjust the boundaries used in the computation and initial pre-alignment.  Manual adjustments are generally used as a last resort possibly limiting acceptable agreement to small regions of interest.

I'm reminded of that old Groucho Marx line: "Who are you going to believe, me or your lying eyes?" CT is, relatively speaking, more spatially fidelitous whereas, relatively, MRI is less so. This fact is why, even after an "auto-match," one can sometimes see obvious niggling anatomic and/or target mismatches when rapidly visually cycling back and forth between CT and MRI images. All points on a CT can never be perfectly matched to all points on an MRI without deformable registration of some sort. So for rigid registration there will always be trade-offs. Ultimately for SRS, the most important region of interest is the target, right? Sometimes you can see a target pretty well on contrasted CT and MRI--if you see a mismatch after automatch, I have always manually adjusted. There are times where the target doesn't show well on the CT, but there may be an obvious landmark very nearby which does show well and there may be a mismatch between this structure and the same structure by MRI; I manually adjust in those situations, also. A good example might be a target very near the optic chiasm. Usually, on high-res, thin slice CTs and MRIs, I can spot the chiasm. If a relatively visually CT-inchoate target is very near the chiasm, I will manually adjust (if necessary) for excellent CT/MR chiasm match e.g. Perhaps more common is lack of confidence about where *exactly* the target is by CT and no good anatomic referent is nearby; then you must of course rely much more or purely on automatch. A manual match seems to me foolhardy then. But to your original question, the biggest source of error in the vast chain of possible errors (ah, the "old days"... which we must surely view in disdain nowadays!) which exist in radiosurgical localization is in my opinion the CT-to-MRI mismatch. There is data that manual adjustment is the best way to correct for that. It's all well and good to have a system with sub-mm isocenter error, yet consider the lillies... and other sources of error, as well. 

Agree with all the above. Contrasted CT radiosurgery sim is imperative, esp. in metastatic cases (and of course CTA will be mandatory in an AVM case e.g.). I feel very confident when I can spot the target well on CT, as at the end of the day this is the dataset upon which we 1) ultimately contour using MR as a guide, 2) calculate dose, and 3) match with CBCT (now my preferred method of frameless stereotaxy) at the machine. We probably didn't map the target so well in the pre-CT/MRI fusion radiosurgery days, so while we have these (very mild) worries about CT/MR mismatch, it's still a world of difference better than what we might have been routinely doing two plus decades ago when CT/MR fusion of any sort was not routine. Where I *do not* manually adjust is at the machine for CT/CBCT match for frameless. The automatch algorithm is great for that. When I am without a 6D couch, I insist on all measured rotational errors to be <1 degree in pitch/yaw/roll, and I'm super insistent on this in single iso/multitarget cases. I did a sort of back of the envelope calc that if the average human head diameter is 8 inches, and thus circumference is about 25 inches or 625 mm, you can sort of picture that (unless you place the isocenter geometrically stupidly) no ROI or target should be more than 4 inches from the iso. Thus if you have a perfect match of iso and target in this ~8 inch sphere (iso at center, target at sphere surface) and do a 1 degree rotation, you now have a 625mm*(1/360)=1.7 mm error. So all things considered, I don't know how you *don't* add some margin to the contoured structure in frameless SRS cases given all the errors we have thus far mentioned. It's enough to keep you up at night, but at the end of the day the proof's in the clinical pudding... and it appears all of us do very well with very high local control rates (and essentially nil toxicity in small brain met cases e.g., of course patient/case selection is important).

If you have a siemens scanner, here are mri protocols that minimize distortion/optimized for radiation planning and fusion

https://usa.healthcare.siemens.com/magnetic-resonance-imaging/magnetom-world/hot-topics/mri-in-radiation-therapy/protocols