Are macroscopic PET textural features representative of microscopic heterogeneity? A preliminary simple and direct comparison between PET and digital autoradiography.

1407 Objectives The objective of the study was to analyse the relationship between macroscopic textural features as measured by pre-clinical PET and microscopic heterogeneity extracted from digital autoradiography (dA). Methods Two nude mice with xenografts of MDA-MB-468 human breast carcinoma were injected with 12 MBq of copper semicarbazone compound [64Cu]-diacetyl-bis(N4-methylthiosemicarbazone) (64Cu-ATSM) which has been used as a tracer for hypoxia imaging. One tumor was excised, freezed and cut up into 10 µm width every 500 µm for the purpose of this work. Sixteen slides were extracted and acquired during 16h using the Siemens Inveon PET system. PET images were reconstructed with two different algorithms: OSEM3D (3 iterations, 16 subsets) and OSEM3D+MAP (6 iterations, 16 subsets) without post-filtering. The same slides were subsequently imaged using a new high sensitivity dA (Beaver, Ai4R, France). PET and dA (scaled to the PET voxels size: 0.78[asterisk]0.78[asterisk]0.8 mm3) images were registered using mutual information. A unique delineation obtained using a level sets segmentation algorithm (Sethian, 1996) was extracted from the PET images and applied on the dA images. Eighteen textural features (TF) were subsequently calculated on both images using the co-occurence matrix (6 TFs), the grey level run length matrix (5 TFs) and the grey level size zone matrix (7 TFs) and compared using a Pearson correlation coefficient. Results Over the 18 TF, only 5 TFs presented a significant correlation (r>0.7; p Conclusions These preliminary results suggest that textural features measured by PET may be representative of microscopic heterogeneity for only a limited number of TFs. The study is ongoing and supplementary data will be presented considering other tumors imaged with other isotopes.