Verte-Box: A Novel Convolutional Neural Network for Fully Automatic Segmentation of Vertebrae in CT Image.

Bing Li,Chuang Liu,Shaoyong Wu,Guangqing Li

doi:10.3390/tomography8010005

Bing Li, Chuang Liu + Show 2 more

Open Access

https://doi.org/10.3390/tomography8010005

Copy DOI

Abstract

Due to the complex shape of the vertebrae and the background containing a lot of interference information, it is difficult to accurately segment the vertebrae from the computed tomography (CT) volume by manual segmentation. This paper proposes a convolutional neural network for vertebrae segmentation, named Verte-Box. Firstly, in order to enhance feature representation and suppress interference information, this paper places a robust attention mechanism on the central processing unit, including a channel attention module and a dual attention module. The channel attention module is used to explore and emphasize the interdependence between channel graphs of low-level features. The dual attention module is used to enhance features along the location and channel dimensions. Secondly, we design a multi-scale convolution block to the network, which can make full use of different combinations of receptive field sizes and significantly improve the network’s perception of the shape and size of the vertebrae. In addition, we connect the rough segmentation prediction maps generated by each feature in the feature box to generate the final fine prediction result. Therefore, the deep supervision network can effectively capture vertebrae information. We evaluated our method on the publicly available dataset of the CSI 2014 Vertebral Segmentation Challenge and achieved a mean Dice similarity coefficient of 92.18 ± 0.45%, an intersection over union of 87.29 ± 0.58%, and a 95% Hausdorff distance of 7.7107 ± 0.5958, outperforming other algorithms.

Highlights

Automatic vertebrae segmentation from medical images is critical for spinal disease diagnosis and treatment, e.g., assessment of spinal deformities, surgical planning, and postoperative assessment; computed tomography (CT) is one of the most commonly used imaging methods in clinical practice [1], and the convolutional neural network has become the best choice for processing such images
All scans are resampled to 1 mm × 1 mm × 1 mm, 14 scans are used as a training set, and the remaining 6 scans are used as a test set
Computed tomography is recognized as a gold standard technique to evaluate spinal disease

Summary

Introduction

Automatic vertebrae segmentation from medical images is critical for spinal disease diagnosis and treatment, e.g., assessment of spinal deformities, surgical planning, and postoperative assessment; computed tomography (CT) is one of the most commonly used imaging methods in clinical practice [1], and the convolutional neural network has become the best choice for processing such images.In practice, vertebrae segmentation from volumetric CT image suffers from the following challenges: Inter-class similarity: Shape and appearance similarities appear in the neighboring vertebrae from the sagittal view. Vertebrae segmentation has been approached predominantly as a model-fitting problem using statistical shape models (SSM) [2,3] or active contour methods [4,5] in the early stage. Chu et al [6] detected the center of the vertebral bodies using random forest and Markov random fields and used these centers to obtain fixed-size regions of interest (ROI), in which vertebrae were segmented using random forest voxel classification. Korez et al [7] used a convolution neural network to generate a probability map of vertebrae, and these maps were used to guide the deformable model to segment each vertebra. Machine learning methods outperform the traditional approaches in speed and efficiency, they have no obvious advantage in segmentation accuracy

Objectives

Methods

Results

Discussion

Conclusion