SU‐FF‐I‐124: Automatic Identification of Water and Fat Images From a Symmetrically‐Sampled Dual‐Echo Dixon Technique

Purpose: Dixon techniques can produce useful separate water‐only and fat‐only images. Correct automatic identification of water and fat images after they are separated is a necessary requirement for proper image annotation and archival. Previously, the difference between the signal intensity histograms of water‐only and fat‐only images has been exploited for image identification. Here we demonstrate a more robust method that is based on the intrinsic asymmetry between the water and fat chemical shift spectra. Method and Materials: A total of 131 patient datasets (each containing approximately 100 separated water‐only and fat‐only image pairs) were randomly selected from different scanners. The proposed method is based on the observation that while water spectrum is a single peak, the fat spectrum is approximately bi‐modal and contains a secondary peak near water resonance. Further, the method exploits the fact that most in vivo tissues are either water‐dominant or fat‐dominant. As a result, pixels with only fat will in general have non‐zero signal intensity in the water‐only image. In contrast, pixels with only water will have zero intensity in the fat‐only image. A simple statistical parameter was calculated from the pair of the water‐only and fat‐only images and used to decide which of the two images is water and which is fat. Visual identification was used as the gold standard. Results: The spectrum based image identification method was correct in 98% of the 13496 total image pairs. In comparison, a pure histogram based algorithm had a success rate of only 41.5%. When applied to only the central 90% of the slices for all image datasets, the new method easily achieved a 100% success rate. Conclusion: In comparison to a pure histogram based method, the new method is shown to be much less susceptible to variations in scanning protocols, patients, and the injection of contrast agents.