A dedicated ML-algorithm for non-attenuation corrected PET whole body images

In whole body positron emission tomography, it is common practice to omit the transmission scan to shorten the study time. Consequently, the emission sinograms are reconstructed without attenuation correction. Although this leads to artifacts in the reconstructed images, it has been shown that these images still provide valuable diagnostic information in oncological applications. Analysis of the attenuation artifact reveals that it consists of two components: a multiplicative and an additive component. The presence of the additive component explains why tumor to background ratio decreases with attenuation correction, a puzzling finding reported by several groups. In cases of strong attenuation, the negative additive component can have an adverse effect on image quality in maximum-likelihood reconstruction. The problem is caused by the non-negativity constraint of that algorithm. The authors propose a new ML-algorithm in which this constraint is removed to preserve image quality. Since the introduction in the clinical routine in the authors´ department, the algorithm has been applied to over 2000 studies with satisfactory results