68Ga-DOTATOC PET radiomics as a novel preoperative predictive tool for PanNETs aggressiveness

#3851

Introduction: The knowledge of Pancreatic Neuroendocrine Tumors (PanNETs) features of aggressiveness before surgery is fundamental for an appropriate treatment planning.

Aim(s): To investigate the preoperative role of 68Ga-DOTATOC PET radiomics in predicting features of PanNETs aggressiveness.

Materials and methods: Retrospective study including 76 PanNET patients who underwent 68Ga-DOTATOC PET for preoperative staging in absence of neoadjuvant treatments. Features of aggressiveness including grade (G), lymph nodes (LN) involvement, presence of liver metastases (LM) and DAXX loss of expression (DAXX LoE) were collected from surgical specimens. PET were segmented by two nuclear medicine physicians and pre-processed. Radiomic features (RFs) were extracted and selected first for intra- and inter-observer reproducibility (ICC >0.75), and then using the mRMR algorithm. For the prediction of each outcome, a machine learning classifier was implemented and validated by means of 50-times repeated 3-fold stratified cross-validation. Averaged area under the curve (AUC), accuracy (ACC), sensitivity (SN) and specificity (SP) were collected.

Conference:

Presenting Author:

Authors: Mapelli P, Bezzi C, Ghezzo S, Muffatti F, Andreasi V,

Keywords: 68Ga-DOTATOC, NET, PET, radiomics,