Fully automated segmentation and lymph node metastasis prediction of non-functional pancreatic neuroendocrine tumours using deep learning

#4379

Introduction: Lymph node status is an important factor for the patients with NF-PanNETs with respect to surgical methods, prognosis, and recurrence. Our model serves as a non-invasive tool that supports clinical decision-making for precision surgical treatment in patients with NF-PanNETs.

Aim(s): We propose to automatically segment and predict lymph node metastasis of non-functional pancreatic neuroendocrine tumours with a deep learning (DL)-based approach on CT imaging.

Materials and methods: In this retrospective study, we collect the arterial phase of the contrast-enhanced CT from 406 patients with NF-PanNETs who underwent surgical resection. The status of LNM was confirmed by surgical pathology. We randomly split the 406 patients into 286/72/48 for the purpose of training/validation/testing. A deep learning-based model was developed to simultaneously segment the tumour and predict the binary status of LNM, supervised by the manual tumour segmentations and LNM standard of truths in the training set. This model was then validated in the validation set and the independent test set, without the need for manual segmentation.

Conference:

Presenting Author: Tang W

Authors: Tang W, Chen J,

Keywords: Non-functional pancreatic neuroendocrine tumour, deep learning, fully automated segmentation, imaging Informatics,