Reconstruction Method for Optical Tomography Based on Generative Adversarial Network

Fig. 5. Five specific distributions. (a) Single rectangle distribution; (b) single circle distribution; (c) double circle distribution; (d) triple circle distribution; (e) cross-distribution

Download full size

Fig. 6. The optimized optical tomographic sensor structure

Download full size

Fig. 7. Trends of equivalent cross entropy. (a) Equivalent cross-entropy of the generator; (b) equivalent cross-entropy of the discriminator

Download full size

Fig. 8. The images reconstructed by a generator in different iterations

Download full size

Fig. 9. Reconstruction images of the four methods. (a) (e) (i) (m) (q) Reconstruction images of LBP; (b) (f) (j) (n) (r) reconstruction images of Landweber; (c) (g) (k) (o) (s) reconstruction images of U-Net; (d) (h) (l) (p) (t) reconstruction images of LBP-Pix2Pix

Download full size

Fig. 10. The reconstruction errors of four methods

Download full size

Table 1. Parameters in training Pix2Pix model
View table
Table 1. Parameters in training Pix2Pix model
Parameter Input image size Crop size Epoch Batch size Learning rate Momentum $β_{1}$ Momentum $β_{2}$ L1 weighted term $λ$
Value 256×256×3 224×224×3 23 1 0.0002 0.5 0.999 100

Table 2. Parameters of different absorption coefficient distributions in the dataset

View table

Table 2. Parameters of different absorption coefficient distributions in the dataset

Absorption coefficient distribution	Random circle		Random rectangle		Random cross
Absorption coefficient distribution	Diameter	Center coordinates	Length and width	The lower-left coordinates	The width of the whole cross	Width of the intersection
Range of parameter	［20，40］	［-30，30］	［10，30］	［-20，20］	［30，40］	［10，20］

Table 3. Parameters of simulation experiment

View table

Table 3. Parameters of simulation experiment

Parameter	Uniformity coefficient $S_{u}$	The divergence angle of the emitter $θ$ /rad	The number of emitters and receivers	The diameter of the region of interest /mm	The number of pixels in the region of interest	The size of each pixel /mm
Value	0.288	1.6	25	100	1860	2

Table 4. E_RIEof other methods
View table
Table 4. E_RIEof other methods
Method Ⅰ Ⅱ Ⅲ Ⅳ Ⅴ $σ$
LBP 0.4586 0.4759 0.3978 0.3418 0.4126 0.0596
Landweber 0.1540 0.1636 0.1644 0.2450 0.3087 0.0680
U-Net 0.1226 0.2213 0.2847 0.3196 0.1533 0.0782
LBP-Pix2Pix 0.0520 0.1037 0.1315 0.1206 0.1177 0.0519

Table 5. C_ICC of different methods
View table
Table 5. C_ICC of different methods
Method Ⅰ Ⅱ Ⅲ Ⅳ Ⅴ $σ$
LBP 0.5174 0.6248 0.4934 0.5588 0.5767 0.0719
Landweber 0.9438 0.9602 0.9329 0.8841 0.8181 0.0778
U-Net 0.9070 0.7948 0.6462 0.6593 0.8785 0.1209
LBP-Pix2Pix 0.9908 0.9698 0.9143 0.8834 0.9127 0.0445

Table 6. E_RSME of different methods
View table
Table 6. E_RSME of different methods
Method Ⅰ Ⅱ Ⅲ Ⅳ Ⅴ $σ$
LBP 0.0562 0.0616 0.0558 0.0717 0.0789 0.0091
Landweber 0.0145 0.0215 0.0267 0.0459 0.0612 0.0172
U-Net 0.0268 0.0349 0.0465 0.0683 0.0453 0.0140
LBP-Pix2Pix 0.0107 0.0136 0.0240 0.0378 0.0367 0.0113

Table 7. R_PSNR of different methods

View table

Table 7. R_PSNR of different methods

Method	Ⅰ	Ⅱ	Ⅲ	Ⅳ	Ⅴ	$σ$
LBP	25.0040	24.2059	25.0623	22.8936	22.0551	1.1886
Landweber	36.7686	33.3620	31.4771	26.7625	24.2705	4.4992
U-Net	31.4284	29.1353	26.6563	23.1279	26.8785	2.7678
LBP-Pix2Pix	39.3869	37.3385	32.4107	28.4517	28.7063	4.4441

Table 8. M_SSIM of different methods
View table
Table 8. M_SSIM of different methods
Method Ⅰ Ⅱ Ⅲ Ⅳ Ⅴ $σ$
LBP 0.5126 0.5176 0.5356 0.5616 0.5523 0.0190
Landweber 0.9481 0.8684 0.8169 0.6416 0.5146 0.1579
U-Net 0.9799 0.9375 0.9034 0.7723 0.8796 0.0698
LBP-Pix2Pix 0.9946 0.9781 0.9486 0.8797 0.8992 0.0443

Tools

Get Citation

Copy Citation Text

Yiting Xu, Huajun Li, Yingkuang Zhu, Lianjie Chen, Youhu Zhang. Reconstruction Method for Optical Tomography Based on Generative Adversarial Network[J]. Laser & Optoelectronics Progress, 2024, 61(12): 1211001

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites