A multi-observation fusion approach for patient follow-up using PET/CT

Allowing the early detection of metabolic changes during treatment, Positron Emission Tomography (PET) is a promising tool for therapy response assessment. A therapeutic response is usually defined as variations of semi-quantitative parameters such as standardized uptake value (SUV) measured in PET scans performed during the treatment. However, this approach does not take into account volume variation and cannot address heterogeneous uptake variations within a tumor volume. Our fusion method is derived from multi-observation approaches principally used in the astronomy field and is aimed at merging several follow-up PET acquisitions in order to assess an accurate tumor metabolic volume variation, by automatically delineating the various uptake and shape variations of the tumor. The statistical approach developed assumes data can be modeled by a mixture distribution of random fields. The parameters defining the mixture distribution are estimated using the stochastic expectation maximization (SEM) method. The proposed fusion process has been applied to simulated follow-up PET images, considering tumors with non spherical shapes and inhomogeneous activity distributions. To assess the relevance of our fusion method, we have compared the resulting fused map with superposition of individual segmented maps obtained with the standard Fuzzy C-Means (FCM) algorithm. In the various cases considered the multi-observation fusion was more efficient than the superposition of FCM maps for the automatic delineation of different concentration activities based on the comparison of volume errors for both approaches. Future work will consist in evaluating the ability of the ASEM fusion method to evaluate clinical response to therapy in oncology applications.