Difference Aware Guided Boundary Transformer Network for Childhood Pneumonia CT Image Segmentation

Jia Lü; Mingkai Yu; Xin Chen; Ling He

doi:10.3788/AOS250760

Acta Optica Sinica, Volume. 45, Issue 15, 1510003(2025)

Difference Aware Guided Boundary Transformer Network for Childhood Pneumonia CT Image Segmentation

Jia Lü^1,2、*, Mingkai Yu¹, Xin Chen³, and Ling He^3、**

Author Affiliations

¹College of Computer and Information Sciences, Chongqing Normal University, Chongqing 401331, China

²National Center for Applied Mathematics in Chongqing, Chongqing Normal University, Chongqing 401331, China

³Intelligent Application of Big Data in Pediatrics Engineering Research Center of Chongqing Education Commission of China, Ministry of Education Key Laboratory of Child Development and Disorders, National Clinical Research Center for Child Health and Disorders, Department of Radiology Children’s Hospital of Chongqing Medical University, Chongqing 400014, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(14)

Fig. 1. Structure of the DBTU-Net

Download full size

Fig. 2. Structure of Gated Channel Transformation

Download full size

Fig. 3. Structure of boundary Transformer module

Download full size

Fig. 4. Sample examples in the Child-P dataset and their corresponding labels

Download full size

Fig. 5. Confusion matrices for the four metrics under different hyperparameter configurations on the Child-P dataset. (a) JI; (b) SE; (c) MCC; (d) HD

Download full size

Fig. 6. Segmentation results (up) and their confidence maps (down) on the Child-P dataset. (a) Label; (b) DBTU-Net; (c) U-Net; (d) U-Net++; (e) TransDeepLab; (f) CASCADE; (g) MEGANet

Download full size

Fig. 7. Segmentation results (up) and their confidence maps (down) on the COVID and MosMed dataset. (a) Label; (b) DBTU-Net; (c) U-Net; (d) U-Net++; (e) TransDeepLab; (f) CASCADE; (g) MEGANet

Download full size

Fig. 8. Local segmentation results and heatmaps of different methods on three datasets. (a) Original images; (b) local label; (c) proposed method; (d) U-Net; (e) CASCADE; (f) MEGANet

Download full size

Fig. 9. Comparison of computational complexity on the Child-P dataset

Download full size

Table 1. Evaluation metrics
View table
Table 1. Evaluation metrics
Metrics Description
Dice /% $2 \times | \hat{y} ⋂ y | / | \hat{y} | + | y |$
JI /% $| \hat{y} ⋂ y | / | \hat{y} ⋃ y |$
SE /% $N_{T P} / (N_{T P} + N_{F N})$
MCC /% $\frac{N_{T P} \times N_{T N} - N_{F P} \times N_{F N}}{{[(N_{T P} + N_{F P}) \times (N_{T P} + N_{F N}) \times (N_{T N} + N_{F P}) \times (N_{T N} + N_{F N})]}^{1 / 2}}$
HD /pixel $m a x \{m a x_{p \in P} \{m i n_{y \in Y} ‖p - y‖\}, m a x_{y \in Y} \{m i n_{p \in P} ‖p - y‖\}\}$

Table 2. Experimental results for different combinations of modules on the Child-P dataset

View table

Table 2. Experimental results for different combinations of modules on the Child-P dataset

Combinations of modules	Dice （↑）	JI （↑）	SE （↑）	MCC （↑）	HD /pixel （↓）
U-Net （32）	0.8048	0.6734	0.7277	0.8066	62.04
U-Net （32）+GCT	0.8802	0.7860	0.8666	0.8784	30.50
U-Net （32）+DAF	0.8869	0.7968	0.8671	0.8854	30.50
U-Net （32）+BT	0.8825	0.7897	0.8676	0.8808	33.17
U-Net （32）+DAF+BT	0.8912	0.8037	0.8796	0.8895	25.24
U-Net （32）+GCT+DAF	0.8920	0.8052	0.8856	0.8904	36.81
U-Net （32）+GCT+BT	0.8894	0.8008	0.8832	0.8876	30.00
DBTU-Net	0.8960	0.8117	0.8931	0.8944	25.08

Table 3. Comparison of experimental results with existing networks on the Child-P dataset

View table

Table 3. Comparison of experimental results with existing networks on the Child-P dataset

Network	Dice	JI	SE	MCC	HD /pixel
U-Net^［7］	0.8144	0.6869	0.7547	0.8142	47.30
ResU-Net^［8］	0.8017	0.6690	0.7301	0.8027	56.82
U-Net++^［9］	0.8394	0.7233	0.7917	0.8385	46.32
TMU-Net^［19］	0.7893	0.6519	0.7145	0.7907	35.44
TransDeepLab^［21］	0.8466	0.7340	0.8317	0.8443	49.50
CSU-Net^［22］	0.8711	0.7716	0.8730	0.8690	58.01
CASCADE^［31］	0.8803	0.7862	0.8789	0.8783	54.64
CHWS-UNet^［32］	0.8525	0.7429	0.8153	0.8511	84.13
MEGANet^［33］	0.8746	0.7771	0.8502	0.8730	34.67
DBTU-Net	0.8961	0.8117	0.8931	0.8944	25.08

Table 4. Comparison of experimental results with existing networks on the COVID dataset

View table

Table 4. Comparison of experimental results with existing networks on the COVID dataset

Network	Dice	JI	SE	MCC	HD /pixel
U-Net^［7］	0.7708	0.6271	0.7139	0.7574	22.56
ResU-Net^［8］	0.7846	0.6455	0.7077	0.7751	22.56
U-Net++^［9］	0.8085	0.6786	0.7429	0.7987	32.39
TMU-Net^［19］	0.7803	0.6397	0.6987	0.7714	28.20
TransDeepLab^［21］	0.7758	0.6337	0.7191	0.7626	35.52
CSU-Net^［22］	0.7734	0.6305	0.7501	0.7569	23.83
CASCADE^［31］	0.8052	0.6739	0.7421	0.7949	22.65
CHWS-UNet^［32］	0.7737	0.6309	0.7351	0.7584	28.64
MEGANet^［33］	0.7972	0.6629	0.7216	0.7883	33.62
DBTU-Net	0.8212	0.6966	0.7702	0.8104	21.93

Table 5. Comparison of experimental results with existing networks on the MosMed dataset

View table

Table 5. Comparison of experimental results with existing networks on the MosMed dataset

Network	Dice	JI	SE	MCC	HD /pixel
U-Net^［7］	0.6983	0.5364	0.6205	0.7016	45.81
ResU-Net^［8］	0.6946	0.5321	0.5942	0.7027	41.68
U-Net++^［9］	0.7115	0.5522	0.6227	0.7168	42.65
TMU-Net^［19］	0.7043	0.5435	0.6630	0.7033	66.71
TransDeepLab^［21］	0.7238	0.5672	0.6866	0.7227	35.83
CSU-Net^［22］	0.6544	0.4863	0.5630	0.6608	36.29
CASCADE^［31］	0.7371	0.5836	0.6628	0.7398	24.29
CHWS-UNet^［32］	0.7008	0.5394	0.5916	0.7112	38.83
MEGANet^［33］	0.7205	0.5631	0.6281	0.7265	43.37
DBTU-Net	0.7437	0.5920	0.6688	0.7465	42.73

Tools

Get Citation

Copy Citation Text

Jia Lü, Mingkai Yu, Xin Chen, Ling He. Difference Aware Guided Boundary Transformer Network for Childhood Pneumonia CT Image Segmentation[J]. Acta Optica Sinica, 2025, 45(15): 1510003

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category: Image Processing

Received: Mar. 18, 2025

Accepted: May. 6, 2025

Published Online: Aug. 8, 2025

The Author Email: Jia Lü (lvjia@cqnu.edu.cn), Ling He (heling508@sina.com)

DOI:10.3788/AOS250760

CSTR:32393.14.AOS250760

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology

Table 1. Evaluation metrics

Table 1. Evaluation metrics

Table 2. Experimental results for different combinations of modules on the Child-P dataset

Table 2. Experimental results for different combinations of modules on the Child-P dataset

Table 3. Comparison of experimental results with existing networks on the Child-P dataset

Table 3. Comparison of experimental results with existing networks on the Child-P dataset

Table 4. Comparison of experimental results with existing networks on the COVID dataset

Table 4. Comparison of experimental results with existing networks on the COVID dataset

Table 5. Comparison of experimental results with existing networks on the MosMed dataset

Table 5. Comparison of experimental results with existing networks on the MosMed dataset