A combined nomogram to predict liver metastasis of pancreatic neuroendocrine tumors: Integrating deep learning radiomics and computational pathology

#3960

Introduction: Pancreatic neuroendocrine tumors (panNETs) are a diverse group of tumors, and liver metastasis is an important prognostic factor.

Aim(s): To develop a nomogram that predicts liver metastasis in panNETs using recent advancements in radiomics, deep learning techniques, and computational pathology.

Materials and methods: This retrospective study enrolled panNETs from FUSCC between 2015 and 2022. Radiomic features were extracted using Pyradiomics with Python (version 3.7), while deep learning features were derived using Convolutional Neural Network (ResNET152). Logistic regression was used to derive the deep learning radiomics (DLR) signature. Additionally, Ki-67 stained slides were scanned into whole-slide images with scanning system (KF-pro-005). The sliding window algorithm automatically evaluated Ki-67 hotspot scores and the spatial colocalization of the Ki-67 parameters using the Morisita-Horn (MH) index. An integrated nomogram model was developed by combining selected clinical factor, DLR score, and pathomics parameter.

Conference:

Presenting Author: Huang D

Authors: Huang D, Tang W, Ma M, Liang Y,

Keywords: Pancreatic neuroendocrine tumor, Liver metastasis, Deep learning radiomics, Computational pathology, Nomogram,