Fully automated pelvic bone segmentation in multiparameteric MRI using a 3D convolutional neural network

Xiang Liu,He Wang,Yingpu Cui,Chao Han,Xiaoying Wang,Jingyun Wu,Xiaodong Zhang

doi:10.1186/s13244-021-01044-z

Abstract

BackgroundAccurate segmentation of pelvic bones is an initial step to achieve accurate detection and localisation of pelvic bone metastases. This study presents a deep learning-based approach for automated segmentation of normal pelvic bony structures in multiparametric magnetic resonance imaging (mpMRI) using a 3D convolutional neural network (CNN).MethodsThis retrospective study included 264 pelvic mpMRI data obtained between 2018 and 2019. The manual annotations of pelvic bony structures (which included lumbar vertebra, sacrococcyx, ilium, acetabulum, femoral head, femoral neck, ischium, and pubis) on diffusion-weighted imaging (DWI) and apparent diffusion coefficient (ADC) images were used to create reference standards. A 3D U-Net CNN was employed for automatic pelvic bone segmentation. Additionally, 60 mpMRI data from 2020 were included and used to evaluate the model externally.ResultsThe CNN achieved a high Dice similarity coefficient (DSC) average in both testing (0.80 [DWI images] and 0.85 [ADC images]) and external (0.79 [DWI images] and 0.84 [ADC images]) validation sets. Pelvic bone volumes measured with manual and CNN-predicted segmentations were highly correlated (R2 value of 0.84–0.97) and in close agreement (mean bias of 2.6–4.5 cm3). A SCORE system was designed to qualitatively evaluate the model for which both testing and external validation sets achieved high scores in terms of both qualitative evaluation and concordance between two readers (ICC = 0.904; 95% confidence interval: 0.871–0.929).ConclusionsA deep learning-based method can achieve automated pelvic bone segmentation on DWI and ADC images with suitable quantitative and qualitative performance.

Highlights

Accurate segmentation of pelvic bones is an initial step to achieve accurate detection and localisation of pelvic bone metastases
A SCORE system was designed for the qualitative evaluation of segmentation
It lays a foundation for the detection of pelvic bony metastases

Summary

Introduction

Accurate segmentation of pelvic bones is an initial step to achieve accurate detection and localisation of pelvic bone metastases. This study presents a deep learning-based approach for automated segmentation of normal pelvic bony structures in multiparametric magnetic resonance imaging (mpMRI) using a 3D convolutional neural network (CNN). Liu et al Insights Imaging (2021) 12:93 intensity) and quantitative (apparent diffusion coefficient [ADC] maps) information for lesion detection and characterisation [2,3,4] Adverse bone events, such as pathological fracture and spinal cord compression, were often led by bone metastases [5, 6]. The initial step to achieve accurate bone metastases detection on DWI and ADC images requires accurate skeleton segmentation with their semantic labels. It is the first step in developing an automated method for quantifying skeletal metastatic tumour burden. Fully convolutional neural networks (CNNs) such as the U-Net model proposed by Ronneberger et al [10] and the V-Net model proposed by Milletari et al [11] have significantly increased the potential of automated image analysis to an unprecedented level

Objectives

Methods

Results

Discussion

Conclusion