The European Association of Nuclear Medicine (EANM) and the European Society of Cardiology (ESC) have brought together their expertise in order to develop guidelines for myocardial perfusion imaging in nuclear cardiology. The result [1] is a very rich text with a lot of highly useful clinical and technical details.

In the eighth section, dealing with reconstruction methods, it is stated that “although FBP is probably still (2004) the most frequently used method of SPECT reconstruction, it is rapidly losing ground to the iterative techniques”. This section gives an excellent introduction to both FBP and MLEM/OSEM reconstruction methods. However, it is our opinion that one parameter of the utmost importance for correct use of OSEM is not well documented.

OSEM [2] can be viewed as a “trick” to accelerate the MLEM algorithm. The trick consists in grouping the projections into small sets to form subsets. Although there are several ways to realise the grouping, commercial algorithms propose only one method and no alternative is available to the user. There are, however, two parameters that should generally be fixed by the user. The first is, as in MLEM, the number of iterations (i). The second is the number of projections per subset, or the equivalent number of subsets (s). As stated in the guidelines [1], the convergence of the OSEM algorithm is not theoretically demonstrated, whereas the convergence of MLEM is. It is, however, a common experience that i OSEM iterations with s subsets deliver images very close to those generated by i×s MLEM iterations [3]. The main advantage is a gain in processing time roughly equal to the number of subsets [1]. Ten years of experience with OSEM have shown that the subsets should contain at least 4 projections. With fewer projections per subset, the OSEM algorithm is more likely to diverge. This means a maximum of 8 subsets for 32 projections, 16 subsets for 64 projections and 32 subsets for 128 projections. The number of subsets is given neither in the guidelines [1], nor in the text, nor in Table 11. In Table 11, 10–15 MLEM iterations and 1–2 OSEM iterations are recommended. From these two sets of data and the above-mentioned rule of equivalence between OSEM and MLEM produced images, it could be inferred that the recommended number of subsets should be 8–16. However, it needs to be borne in mind that more than 8 subsets are not suitable for 32 projections.

In summary, it could be recommended that the inexperienced user should always use 8 subsets and at least 2 iterations for OSEM reconstruction of SPECT myocardial perfusion images. The user should nevertheless also be made aware that such a low number of iterations will generate reconstructed images with a low noise level, but at the expense of a biased contrast [4]. In other words, a low number of iterations results in smooth images, but with a low contrast, especially in poorly perfused regions [4, 5].

Furthermore, in light of the ever-increasing computer performance, it could be questioned whether OSEM should still be used in SPECT when no correction (attenuation, scatter, resolution) is performed.