Texture analysis in metastases of patients affected by metastatic differentiated thyroid carcinoma treated with 131I

In metastatic differentiated thyroid cancer (MDTC) the radioiodine treatment is a consolidated approach. The effectiveness of the therapy can be verified on post administration tomography imaging. Aim of this study is to evaluate the feasibility and prognostic impact of 131I SPECT-CT images features analysis (FA) performed on metastatic lesion in MDTC patients. FA is performed on SPECT-CT images with a free software (LIFEx, CEA 2017) taking into account texture features (TF) and first-order features (FF). TF include Gray-Level Occurrence Matrix GLCM (e.g. homogeneity, contrast) and Gray-Level Zone Length Matrix GLZLM (e.g. HGZE); FF include histograms indices (e.g. skewness, kurtosis and entropy). On a spheres phantom (NEMA PET IEC) filled with liquid 131I (13 MBq/ml), SPECT-CT images (at 3, 120, 160, 216 h after preparation) were acquired. Five volume of interests (VOIs) were contoured on phantom images, one for each sphere. VOIs were made by a 3D segmentation tool choosing a threshold value to obtain a volume the most similar to the real one. Then, we performed TF and FF analysis using six different discretization levels (4, 8, 16, 32, 64, 128) on every single VOI. As results, the best discretization level to use and the minimum VOI volume that should be evaluated with this method were defined. For 8 MDTC patients (inclusion criteria: thyroidectomy, iodine remnant ablation) TF and FF analysis were computed on 4 SPECT-CT images (Siemens IntevoT2, 256x256, ITSCAC) acquired at 4, 24, 48, 96 h after administration. Mean doses to lesion (MIRD sphere model) were also evaluated. 51 lesions (37 bone, 14 visceral) were contoured by a nuclear medicine specialist using a 3D segmentation tool and TF and FF analysis were computed. Treatment outcome was classified as responding (complete and partial response) or not responding (stable and progressive disease). FA was therefore correlated to 131I treatment response through ROC analysis. In phantom analysis the minimal assessable volume was found equal to 1.2 cc and the optimal discretization level was found equal to 64, as suggested by literature: those results were therefore applied in patients analysis. In patients, mean dose ROC analysis with p value equal to 81,0 confirm the literature data which show 80 Gy as a threshold above which there is a greater percentage of a good treatment response. Similar results were found on some GLCM features (homogeneity and contrast). Higher effectiveness to discriminate the patient treatment response was shown by some FF features (skewness, kurtosis and entropy) as well by TF Gray-Level Zone Length Matrix In MDTC patients features analysis on 131I SPECT-CT images is feasible. Phantom analysis points out methods limits, allowing a standardization. Many features were evaluated in order to investigate their influence in the patient response discrimination. A positive prognostic capability was shown. These results should be integrated with the standard dose-response evaluation, giving to the physician a powerful tool in MDTC patients treatment management. However, the small sample involved does not allow clinical results yet: additional patients involved would consolidate the prognostic and predictive role of FA on 131I treatment response.