Automatic Histogram Specification for Glioma Grading Using Multicenter Data.

Xi Chen,Wenyi Gao,Guohua Zhao,Yaping Wu,Yusong Lin,Qian Zhang,Meiyun Wang

doi:10.1155/2019/9414937

Abstract

Multicenter sharing is an effective method to increase the data size for glioma research, but the data inconsistency among different institutions hindered the efficiency. This paper proposes a histogram specification with automatic selection of reference frames for magnetic resonance images to alleviate this problem (HSASR). The selection of reference frames is automatically performed by an optimized grid search strategy with coarse and fine search. The search range is firstly narrowed by coarse search of intraglioma samples, and then the suitable reference frame in histogram is selected by fine search within the sample selected by coarse search. Validation experiments are conducted on two datasets GliomaHPPH2018 and BraTS2017 to perform glioma grading. The results demonstrate the high performance of the proposed method. On the mixed dataset, the average AUC, accuracy, sensitivity, and specificity are 0.9786, 94.13%, 94.64%, and 93.00%, respectively. It is about 15% higher on all indicators compared with those without HSASR and has a slight advantage over the result of a manually selected reference frame by radiologists. Results show that our methods can effectively alleviate multicenter data inconsistencies and lift the performance of the prediction model.

Highlights

Glioma is a prevalent fatal brain disease and the most malignant, which accounts for approximately 24.7% of all primary brain and other central nervous system tumors and 74.6% of malignant tumors [1]. e World Health Organization’s guidelines for glioma diagnosis and treatment are divided into four levels, namely, I–II and III–IV for lowgrade glioma (LGG) and high-grade glioma (HGG) [2]
Dataset. ese methods include the histogram equalization (HE) method, reference frame method of manual selection by radiologists, and the method proposed in this paper. e results show that compared with the other two methods, the method proposed in this paper significantly improves the grading effect of glioma. e reference frame selected by radiologists based on experience shows significant improvement in glioma grading, but it can be seen from the results that the performance of the selected reference frame is worser than that of the automatically selected
Inconsistencies among data prevent multicenter data from playing to its shared advantage. is paper proposes a histogram specification method with automatic selection of reference frames for magnetic resonance images to alleviate the problem of contrast inconsistencies among multicenter data. e core of histogram specification is to change the local brightness of the image according to the reference frame, but the reference frame of traditional histogram specification is usually manually selected by radiologists

Summary

Introduction

Glioma is a prevalent fatal brain disease and the most malignant, which accounts for approximately 24.7% of all primary brain and other central nervous system tumors and 74.6% of malignant tumors [1]. e World Health Organization’s guidelines for glioma diagnosis and treatment are divided into four levels, namely, I–II and III–IV for lowgrade glioma (LGG) and high-grade glioma (HGG) [2]. E World Health Organization’s guidelines for glioma diagnosis and treatment are divided into four levels, namely, I–II and III–IV for lowgrade glioma (LGG) and high-grade glioma (HGG) [2]. Biological behavior, treatment options, and prognoses of patients with glioma of different grades are clearly different. Erefore, the accurate preoperation grading of Glioma is important. Magnetic Resonance Imaging (MRI) is characterized by multidirectional tomography and multiparameter high-resolution soft-tissue imaging and is widely used to evaluate the tumor heterogeneity [3]. MRI is commonly used in glioma grading because it can accurately display the location and size, and it correlates well with histological characteristics. High quality data is difficult to obtain in a single institution, so it needs to be shared through multicenter. The difference of multicentre data is a serious challenge

Methods

Results

Conclusion