End-to-End Segmentation of Brain White Matter Hyperintensities Combining Attention and Inception Modules

Xin Zhao; Xin Wang; Hongkai Wang

doi:10.3788/AOS202141.0910002

Journals >Acta Optica Sinica >Volume 41 >Issue 9 >Page 0910002 > Article

Acta Optica Sinica
Vol. 41, Issue 9, 0910002 (2021)

End-to-End Segmentation of Brain White Matter Hyperintensities Combining Attention and Inception Modules

Xin Zhao^1、*, Xin Wang¹, and Hongkai Wang²

Author Affiliations

¹School of Information Engineering, Dalian University, Dalian, Liaoning 116622, China

²School of Biomedical Engineering, Dalian University of Technology, Dalian, Liaoning 116024, China

show less

DOI: 10.3788/AOS202141.0910002 Cite this Article Set citation alerts

Xin Zhao, Xin Wang, Hongkai Wang. End-to-End Segmentation of Brain White Matter Hyperintensities Combining Attention and Inception Modules[J]. Acta Optica Sinica, 2021, 41(9): 0910002 Copy Citation Text

show less

Fig. 1. Structure of overall model

Download full size

Fig. 2. Inception module. (a) Original Inception-ResNet-A module; (b) improved module

Download full size

Fig. 3. Optimization process of model based on CBAM

Download full size

Fig. 4. Segmentation results of different improved models. (a) Ground truth; (b) UD; (c) UDIA; (d) UTIA

Download full size

Fig. 5. Comparison of segmentation results between UTIA and Res-UNet. (a) GT; (b) ResU-Net; (c) UTIA

Download full size

Fig. 6. Comparison of loss function curves under different conditions. (a) Model with residual connections; (b) model with no residual connections

Download full size

Model	Recall	Precision	DSC
UD	0.65	0.82	0.72
UDI^*	0.63	0.77	0.74
UDI	0.67	0.85	0.75
UDA	0.70	0.84	0.76
UDIA	0.74	0.79	0.76
UT	0.69	0.77	0.72
UTI	0.76	0.74	0.75
UTA	0.74	0.79	0.76
UTIA	0.84	0.71	0.77

Table 1. Comparison of evaluation indexes of different improved models

Method	Model	Recall	Precision	DSC
Ref. [7]	Random forest	0.27	0.29	0.50
Ref. [12]	MD-GRU	0.81	0.69	0.75
Ref. [16]	Ensemble U-Net	0.84	0.76	0.80
Ref. [17]	ResU-Net	0.83	0.70	0.78
Ref. [19]	Multi-scale U-Net	0.86	0.76	0.80
Ref. [21]	SC-UNet	0.81	0.70	0.78
Proposed method	UTIA	0.84	0.71	0.77

Table 2. Comparison of evaluation indexes between proposed method and other methods

Xin Zhao, Xin Wang, Hongkai Wang. End-to-End Segmentation of Brain White Matter Hyperintensities Combining Attention and Inception Modules[J]. Acta Optica Sinica, 2021, 41(9): 0910002

Download Citation

Set citation alerts for the article

Tools

Set citation alerts for the article

Save the article for my favorites

Paper Information