Three-Dimensional Segmentation of Brain Tumors in Magnetic Resonance Imaging Based on Improved Continuous Max-Flow

Lu Ren; Qiang Li; Xin Guan; Jie Ma

doi:10.3788/LOP55.111011

Laser & Optoelectronics Progress, Volume. 55, Issue 11, 111011(2018)

Three-Dimensional Segmentation of Brain Tumors in Magnetic Resonance Imaging Based on Improved Continuous Max-Flow

Lu Ren¹, Qiang Li^1、*, Xin Guan¹, and Jie Ma²

Author Affiliations

¹ School of Microelectronics, Tianjin University, Tianjin 300072, China

² Tianjin Weishen Technology Company Limited, Tianjin 300384, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(12)

Fig. 1. Framework of the segmentation algorithm proposed

Download full size

Fig. 2. Four types of brain tumor MRI images and the expert segmentation result

Download full size

Fig. 3. Three types of brain tumor MRI images and the fused image. (a) Flair; (b) T1C; (c) T2; (d) fused image

Download full size

Fig. 4. Histograms of three types of brain tumor MRI images and the fused image. (a) Flair; (b) T1C; (c) T2; (d) fused image

Download full size

Fig. 5. Main steps of HGG 3D segmentation. (a) Under-segmented image; (b) accurate-segmented image; (c) final segmented result; (d) gold standard

Download full size

Fig. 6. Main steps of LGG 3D segmentation. (a) Under-segmented image; (b) accurate-segmented image; (c) final segmented result; (d) gold standard

Download full size

Table 1. Processing result of 45 images for different ratios

View table

Table 1. Processing result of 45 images for different ratios

Fusion ratio	1∶0∶0	0∶1∶0	0∶0∶1	8∶0∶2	7∶1∶2	7∶0∶3	6∶0∶4	6∶1∶3	5∶2∶3	5∶1∶4
Dice	0.8277	0.2066	0.6989	0.8329	0.8368	0.8502	0.8600	0.8619	0.8608	0.8958
Precision	0.8965	0.4879	0.7687	0.9316	0.9349	0.9269	0.9246	0.9078	0.9014	0.9359
Recall	0.7545	0.3786	0.5758	0.7671	0.7820	0.8062	0.8004	0.8445	0.8601	0.8626

Table 2. Segmentation of the HGG in various directions
View table
Table 2. Segmentation of the HGG in various directions
Direction Under-segmented image Accurate-segmented image Final segmented image Gold standard
Horizontalplane
Coronalplane
Sagittalplane
Dice 0.8645 0.9007 0.9449

Table 3. Segmentation of the LGG in various directions
View table
Table 3. Segmentation of the LGG in various directions
Direction Under-segmented image Accurate-segmented image Final segmented image Gold standard
Horizontalplane
Coronalplane
Sagittalplane
Dice 0.8662 0.9024 0.9038

Table 4. Segmentation performance evaluation of improved FCM segmentation method
View table
Table 4. Segmentation performance evaluation of improved FCM segmentation method
Method Dice Precision Recall Time /min
FCM 0.88 0.92 0.83 0.4
Proposed 0.90 0.94 0.86 0.3

Table 5. Index statistics of HGG, LGG and all data

View table

Table 5. Index statistics of HGG, LGG and all data

Index	Statistics	HGG	LGG	All data
Dice	Max value	0.9449	0.9038	0.9449
	Min value	0.8521	0.8553	0.8521
	Mean	0.8959	0.8848	0.8948
	Standard deviation	0.0251	0.0221	0.0240
Precision	Max value	0.9861	0.9922	0.9922
	Min value	0.8201	0.8034	0.8034
	Mean	0.9359	0.9268	0.9351
	Standard deviation	0.0470	0.0733	0.0492
Recall	Max value	0.9787	0.9412	0.9787
	Min value	0.7582	0.7516	0.7582
	Mean	0.8627	0.8543	0.8619
	Standard deviation	0.0495	0.0681	0.0508

Table 6. Segmentation performance evaluation of three segmentation methods
View table
Table 6. Segmentation performance evaluation of three segmentation methods
Method Dice Precision Recall Time /min
Zhao et al.^[9] 0.87 0.93 0.86 3
Pereria et al.^[10] 0.88 0.89 0.88 7.5
Proposed 0.90 0.94 0.86 0.3

Tools

Get Citation

Copy Citation Text

Lu Ren, Qiang Li, Xin Guan, Jie Ma. Three-Dimensional Segmentation of Brain Tumors in Magnetic Resonance Imaging Based on Improved Continuous Max-Flow[J]. Laser & Optoelectronics Progress, 2018, 55(11): 111011

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites