Texture analysis in 68Ga-DOTATOC PET images for Neuroendocrine Tumors differentiation: a feasibility study.

Neuroendocrine tumors are a quite rare category of tumors, corresponding to about the 0.5% of total malignant tumors. They origin from neuroendocrine system’s cell, which are widely diffused in the organism. In the last decades, the number of diagnosed NET registered a significant increase partly due to the improvement on imaging technologies, including Computed Tomography and Positron Emission Tomography. The thesis is focused on PET images analysis related to 68Ga-DOTATOC radiopharmaceutical, a recetptor tracer analogous to somatostain. The aim of the elaborate consists in the extraction of textural parameters from PET images, in the evaluation of their robustness, and the identification, among them, those that are more related to Ki67. The sample is composed by 42 NET patients: 34 patients with grade G1 tumor, 7 patients with grade G2 and only one patient with a grade G3 tumor. For each of them images from CT and 68Ga-DOTATOC PET are available together with a manual Volume of Interest (VOI) defined by the specialist, using LIFEx support software. From the manual VOI, 4 alternative VOIs are designed using a thresholding method. The Texture Analysis is performed by varyng the type of resampling (relative and absolute), and the number of resampling grey levels (16 32 and 64), for each of the 5 different VOIs. We obtained 8 First Order Features, directly derived from the SUV value of the VOI; and some Second and Higher Order Features: 15 from the Gray Level Co-occurrence Matrix, 14 from the Grey Level Run Length Matrix , 14 from the Grey Level Zone Size Matrix and 5 from the Neighborhood Grey-Level Difference Matrix. To evaluate the robustness of the descriptors, the coefficient of variation, COV, and the p-value from ANOVA was observed. Subsequently, the list of the features that are at the same time robust and correlated to Ki67: Skewness, Kurtosis, Entropy, Energy from FOS; Inverse Different Moment, Sum Entropy, Entropy from GLCM; Short Runs Emphasis, Long Runs Emphasis, Runs Percentage from GLRLM and Large Zone High Grey Level Emphasis from GLZSM. This thesis aims to act as a starting point for future extentions in order to make possible the implementation of a supporting tool able to assist doctor in classifying NET, in their three grades (G1, G2, G3), starting from the Textur Analysis.