Deep learning network for medical volume data segmentation based on multi axial plane fusion

Highlights

• A network segmentation is presented that can extract the features of different pinacoidal planes based on the medical volume data.

• The network is used for the multi-input semantic convolution neural network for fusion and segmentation.

• The global feature maps of different pinacoidal planes could bring further compensation for losses in the segmentation process, so as to obtain a better result.

Abstract

Background and Objective: High-dimensional data generally contains more accurate information for medical image, e.g., computerized tomography (CT) data can depict the three dimensional structure of organs more precisely. However, the data in high-dimension often needs enormous computation and has high memory requirements in the deep learning convolution networks, while dimensional reduction usually leads to performance degradation. Methods: In this paper, a two-dimensional deep learning segmentation network was proposed for medical volume data based on multi-pinacoidal plane fusion to cover more information under the control of computation.This approach has conducive compatibility while using the model proposed to extract the global information between different inputs layers. Results: Our approach has worked in different backbone network. Using the approach, DeepUnet’s Dice coefficient (Dice) and Positive Predictive Value (PPV) are 0.883 and 0.982 showing the satisfied progress. Various backbones can enjoy the profit of the method. Conclusions: Through the comparison of different backbones, it can be found that the proposed network with multi-pinacoidal plane fusion can achieve better results both quantitively and qualitatively.

Introduction

In the resection of gastric cancer, the direction of celiac trunk is particularly important for the surgical plans. Therefore, the segmentation of celiac trunk is of great significance for the diagnosis of gastric surgery. Medical equipment often generate three-dimensional or even four-dimensional data, but due to the limitation of preservation, this data usually need to be formed into two-dimensional or three-dimensional structure. Hence, how to segment more accurate results from the transformed medical data [1], [2]–3] are still one of the most popular topics in the medical image segmentation convolution network [4,5] field.

Due to the complexity of the shape of abdominal organ [6], [7]–8], two dimensional data generally can not display the real shape of the organ in an all round way. In computerized tomography(CT) imaging, the data are expressed as three-dimensional within a broader range thus it can retain more values, and present the exact shape of complex organs more comprehensively. In the research of convolution network segmentation, the use of three-dimensional data will bring tremendous amount of calculation so many researchers tend to transform the data into two-dimension and use other ways to supplement the missing information of the formatting transition. Definitely, this method will bring serious loss of dimensional reduction to the data, and the final segmentation result will naturally be much worse than that of 3D input/output method. Even though many convolutional neural networks can improve the segmentation accuracy to some extent by using feature reusing [9], [10]–11] and other network layer modified algorithms from one pinacoidal of the data [12], [13]–14], they can not supply the original and nondestructive data. Therefore, it is not a good solution to directly use the two-dimensional data of one pinacoidal. It remains a huge progress to be made.

In this paper, we proposed a novel network which can extract the features of different pinacoidal planes based on the medical volume data. The new network is different from the previous methods of convolutional neural network segmentation of one pinacoidal of the data. And then inputing them into the multi-input semantic convolution neural network for fusion and segmentation. The global feature maps of different pinacoidal planes could bring further compensation for losses in the segmentation process. The aim is to obtain a better result.

In general, our contributions are summarized as below:

• We proposed a segmentation structure which could be adapted into many different convolutional network. Moreover, it has hybrid architecture which is composed of multiple branches so as to extract the global information of different pinacoidal planes to make the segmentation of the target more accurate.

• Furthermore, We added X-module part to the network in order to extract the global information between different inputs layers. In the experiment, the feature maps generated by X-module can be merged into the corresponding layer of decoder part in parallel. So it can obtain more information.