Lateral Spine Landmark Detection Based on Matching Clue Regression

Fig. 4. Comparison results between other single-stage methods and proposed method. (a) Input image; (b) results by Zhang et al.^[13]proposed method; (c) results by Ao et al.^[11] proposed method; (d) results by Yi et al.^[9] proposed method; (e) results by our method

Download full size

Fig. 5. Ablation experiment results. (a) Baseline; (b) with weighted loss; (c) with weighted loss and GFA module

Download full size

Table 1. Comparative experiment results between two-stage methods and the proposed method

View table

Table 1. Comparative experiment results between two-stage methods and the proposed method

Method	E_dec	Scaled MAE	MSE
Khanal et al.^［4］ proposed method	17.76	0.007407	0.000235
Chen et al.^［6］ proposed method	23.36	0.007580	0.000472
Zhang et al.^［5］ proposed method	11.01	0.004187	0.000122
Ours	7.14	0.003386	0.000060

Table 2. Comparative experiment results between one-stage methods and the proposed method

View table

Table 2. Comparative experiment results between one-stage methods and the proposed method

Method	E_dec	Scaled MAE	MSE
Yi et al.^［9］ proposed method	13.89	0.005965	0.000321
Zhang et al.^［13］ proposed method	23.83	0.009693	0.001639
Ao et al.^［11］ proposed method	48.84	0.018596	0.025107
Ours	8.84	0.004147	0.000111

Table 3. Comparison results of SMAPE between the proposed method with other advanced methods in TK and LL estimation

View table

Table 3. Comparison results of SMAPE between the proposed method with other advanced methods in TK and LL estimation

Method type	Method	TK /%	LL /%
Two-stage	Khanal et al.^［4］proposed method	37.22	36.32
	Chen et al.^［6］proposed method	22.79	19.45
	Zhang et al.^［5］proposed method	12.83	10.53
One-stage	Yi et al.^［9］proposed method	8.42	5.91
	Zhang et al.^［13］proposed method	8.63	5.55
	Ao et al.^［11］proposed method	17.53	9.49
	Ours	8.40	3.62

Table 4. Ablation experiment results of the proposed method
View table
Table 4. Ablation experiment results of the proposed method
Baseline Weighted loss GFA Center Center+offset Landmark
$\sqrt$ 9.63 14.03 11.75
$\sqrt$ $\sqrt$ 8.28 12.65 10.86
$\sqrt$ $\sqrt$ 6.45 10.73 9.60
$\sqrt$ $\sqrt$ $\sqrt$ 6.21 10.12 8.84

Table 5. Comparison results between proposed GFA module with deformable convolution
View table
Table 5. Comparison results between proposed GFA module with deformable convolution
Method Center Center+offset Landmark
DCN 8.56 12.12 10.68
GFA 6.45 10.73 9.60

Table 6. Validity of landmark matching method
View table
Table 6. Validity of landmark matching method
Method Landmark
w/o matching clue 12.20
w/ matching clue 8.84

Tools

Get Citation

Copy Citation Text

Menghao Gao, Lijun Guo, Rong Zhang, Lixin Ni, Qiang Wang, Xiuchao He. Lateral Spine Landmark Detection Based on Matching Clue Regression[J]. Laser & Optoelectronics Progress, 2024, 61(4): 0417001

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category: Medical Optics and Biotechnology

Received: Apr. 25, 2023

Accepted: May. 29, 2023

Published Online: Feb. 26, 2024

The Author Email: Lijun Guo (guolijun@nbu.edu.cn)

DOI:10.3788/LOP231172

CSTR:32186.14.LOP231172

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology

Table 1. Comparative experiment results between two-stage methods and the proposed method

Table 1. Comparative experiment results between two-stage methods and the proposed method

Table 2. Comparative experiment results between one-stage methods and the proposed method

Table 2. Comparative experiment results between one-stage methods and the proposed method

Table 3. Comparison results of SMAPE between the proposed method with other advanced methods in TK and LL estimation

Table 3. Comparison results of SMAPE between the proposed method with other advanced methods in TK and LL estimation

Table 4. Ablation experiment results of the proposed method

Table 4. Ablation experiment results of the proposed method

Table 5. Comparison results between proposed GFA module with deformable convolution

Table 5. Comparison results between proposed GFA module with deformable convolution

Table 6. Validity of landmark matching method

Table 6. Validity of landmark matching method