The use of deep learning models to predict progression-free survival in patients with neuroendocrine tumors: Results from phase 3 of the RAISE project

#3229

Introduction: Response Evaluation Criteria in Solid Tumors (RECIST) assesses treatment response via tumor progression in patients with neuroendocrine tumors (NET). RAISE combined deep learning models (DLM) with sum of the longest diameter (SLD) of liver lesions and chromogranin A (CgA) to find a surrogate endpoint for RECIST to early predict progression-free survival (PFS).

Aim(s): To investigate early prediction of PFS in patients with NET using DLM.

Materials and methods: Data were taken from CLARINET, a placebo-controlled study assessing lanreotide in tumor control of gastroenteropancreatic NET. A linear Cox model assessed DLM value in improving PFS prediction based on SLD ratio (SLDr) and change from baseline in logCgA. DLM extracted features from lesion-only and whole-liver images from 1690 computerised tomography scans in the CLARINET patient subset with liver lesions and scans for multiple visits over 96 weeks (n=138/204). Lesion and binary mask model comparison assessed features captured.

Conference: 18th Annual ENETS Concerence (2021)

Presenting Author: Pavel M

Authors: Pavel M, Dromain C, Ronot M, Schaefer N, Mandair D,

Keywords: neuroendocrine tumor, deep learning models, progression-free survival, prediction,