Image segmentation of plexiform neurofibromas from a deep neural network using multiple b-value diffusion data

Chang Y Ho,Paul R Territo,Stephen F Kralik,John M Kindler,Kent A Robertson,Scott Persohn

doi:10.1038/s41598-020-74920-1

Abstract

We assessed the accuracy of semi-automated tumor volume maps of plexiform neurofibroma (PN) generated by a deep neural network, compared to manual segmentation using diffusion weighted imaging (DWI) data. NF1 Patients were recruited from a phase II clinical trial for the treatment of PN. Multiple b-value DWI was imaged over the largest PN. All DWI datasets were registered and intensity normalized prior to segmentation with a multi-spectral neural network classifier (MSNN). Manual volumes of PN were performed on 3D-T2 images registered to diffusion images and compared to MSNN volumes with the Sørensen-Dice coefficient. Intravoxel incoherent motion (IVIM) parameters were calculated from resulting volumes. 35 MRI scans were included from 14 subjects. Sørensen-Dice coefficient between the semi-automated and manual segmentation was 0.77 ± 0.016. Perfusion fraction (f) was significantly higher for tumor versus normal tissue (0.47 ± 0.42 vs. 0.30 ± 0.22, p = 0.02), similarly, true diffusion (D) was significantly higher for PN tumor versus normal (0.0018 ± 0.0003 vs. 0.0012 ± 0.0002, p < 0.0001). By contrast, the pseudodiffusion coefficient (D*) was significantly lower for PN tumor versus normal (0.024 ± 0.01 vs. 0.031 ± 0.005, p < 0.0001). Volumes generated by a neural network from multiple diffusion data on PNs demonstrated good correlation with manual volumes. IVIM analysis of multiple b-value diffusion data demonstrates significant differences between PN and normal tissue.

Highlights

We assessed the accuracy of semi-automated tumor volume maps of plexiform neurofibroma (PN) generated by a deep neural network, compared to manual segmentation using diffusion weighted imaging (DWI) data
Post-segmentation, intravoxel incoherent motion (IVIM) parameters generated from the multiple b-value modeling of data are compared between PN and background normal tissue
A total of 35 MRI scans were included from 14 subjects

Summary

Introduction

We assessed the accuracy of semi-automated tumor volume maps of plexiform neurofibroma (PN) generated by a deep neural network, compared to manual segmentation using diffusion weighted imaging (DWI) data. A prototypical NF1 tumor is the plexiform neurofibroma (PN), a neurofibroma variant with tumor cells that spread along multiple nerve fascicles, resulting in an extensive mass of thickened nerve bundles in a proteinaceous matrix These PNs affect 25–50% of NF1 patients and can occur anywhere there are nerve fibers leading to significant morbidity and mortality depending on the size and location adjacent to vital structures[3]. Automated segmentation of PNs from normal tissue with limited user interaction would decrease intra- and interobserver variability as well as user fatigue The purpose of this project is to develop a neural network deep learning algorithm to perform semi-automated volume segmentation of PNs based on multiple b-value diffusion weighted MRI and to assess the accuracy of the tumor volume maps by comparing to manual segmentation. Post-segmentation, intravoxel incoherent motion (IVIM) parameters generated from the multiple b-value modeling of data are compared between PN and background normal tissue

Objectives

Methods

Results

Discussion

Conclusion