Co-analysis of textural pattern and SUVpeak in F18-FDOPA PET/CT significantly improves glioma staging

327 Objectives We studied whether the characterization of tumor texture in FDOPA PET/CT could assist in the identification of tumor grades in both primitive and recurrent gliomas. Methods 81 patients (age: 49±13) with gliomas were studied, including 52 newly-diagnosed tumors and 29 recurrent tumors. Each patient underwent a 40 min dynamic FDOPA PET (Philips Gemini TF) 3-min post injection. Tumors were segmented using a thresholding method accounting for the background activity. For each tumor, the SUVpeak and metabolic volume (MV) were measured, as well as 36 textural indices (TI) obtained using an original normalization approach. The ability of SUVpeak, MV and TI to identify the tumor grade was investigated by using each index alone first (with ROC analyses), and then by using couples consisting of 1 TI with SUVpeak in a binomial model (with ROC analyses and a reclassification method). All patients underwent resection or biopsy and the pathological examination provided what was assumed to be the gold standard grade. Results The 52 newly-diagnosed tumors consisted of 29 high-grade tumors (HG) and 23 low-grade tumors (LG), while the 29 recurrent tumors included 21 HG and 8 LG. Neither SUVpeak nor MV could discriminate LG from HG in newly-diagnosed tumors, while SUVpeak alone could discriminate LG from HG in recurrent tumors (p=0.02). Combining a TI (eg, homogeneity, entropy, or short-run emphasis SRE) with SUVpeak led to a significant LG / HG discrimination for newly-diagnosed tumors (p = 0.01). Among all TI, entropy led to the best reclassification performance: by accounting for both tumor SUVpeak and entropy, 67% of the newly-diagnosed tumors were assigned the correct grade vs 58% when using SUVpeak alone. Combining TI and SUVpeak did not significantly improve the classification of recurrent tumors (86% of properly classified tumors with SUVpeak only). Conclusions The co-analysis of FDOPA-PET SUVpeak and well-selected TI (such as entropy, homogeneity or SRE) made it possible to improve the classification of newly-diagnosed gliomas.