Semantic Segmentation and Diagnosis of Laryngopharyngeal Reflux Based on Class-Balanced Loss

Baozhi Zheng; Houde Dai; Penghua Liu; Hanchen Yao; Zengwei Wang

doi:10.3788/LOP221902

Laser & Optoelectronics Progress, Volume. 60, Issue 14, 1410020(2023)

Semantic Segmentation and Diagnosis of Laryngopharyngeal Reflux Based on Class-Balanced Loss

Baozhi Zheng^1,2, Houde Dai², Penghua Liu², Hanchen Yao², and Zengwei Wang^2、*

Author Affiliations

¹College of Electrical Engineering and Automation, Fuzhou University, Fuzhou 350108, Fujian, China

²Quanzhou Institute of Equipment Manufacturing, Chinese Academy of Sciences, Quanzhou 362216, Fujian, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(8)

Fig. 1. Overall flow chart

Download full size

Fig. 2. Comparison of model predictions under different loss functions. (a) Original images; (b) ground truth; (c) CE-FCN segmentation results; (d) Focal-FCN results; (e) WCE-FCN results; (f) CBD-FCN results

Download full size

Fig. 3. Noise addition and linear regression models

Download full size

Fig. 4. Under threefold cross-validation, the difference between the output total score and the real total score of linear regression

Download full size

Table 1. IoU under different loss functions unit: %

View table

Table 1. IoU under different loss functions unit: %

Function	PC	LV	VF	SG	ARY	GRA	MUC	Mean
CE	76.23	67.67	81.89	38.54	87.44	43.64		56.49
WCE	58.58	47.13	72.14	16.09	81.86	30.25	7.49	44.79
Focal	68.15	50.62	73.55	14.61	80.51	43.76	0.58	47.40
Proposed function	67.20	55.96	77.83	25.16	85.99	50.25	14.08	53.78

Table 2. Indexes of evaluation for recognizing small target areas with various loss functions
View table
Table 2. Indexes of evaluation for recognizing small target areas with various loss functions
Function SG GRA MUC
F1 Specificity F1 Specificity F1 Specificity
CE 0.86 0.92 0.65 0.99 0 1.00
WCE 0.80 0.23 0.76 0.95 0.53 0.73
Focal 0.02 1.00 0.78 0.99 0.22 1.00
Proposed function 0.90 0.75 0.87 0.99 0.75 0.90

Table 3. Results of ablation experiments

View table

Table 3. Results of ablation experiments

Function	PC	LV	VF	SG	ARY	GRA	MUC	Mean
CE	77.45	68.38	82.41	45.80	87.38	48.53	0	58.56
WCE	64.31	51.66	74.24	21.13	83.53	41.64	11.91	49.77
Focal	71.20	52.19	75.44	14.34	83.20	50.49	0.82	49.67
Proposed function	67.98	52.16	76.60	24.33	85.74	42.63	13.22	51.81

Table 4. Three-fold cross validation of the accuracy of LPRD reflux assessment
View table
Table 4. Three-fold cross validation of the accuracy of LPRD reflux assessment
Folds 1 2 3 Average
Accuracy /% 94.73 94.19 94.28 94.40

Tools

Get Citation

Copy Citation Text

Baozhi Zheng, Houde Dai, Penghua Liu, Hanchen Yao, Zengwei Wang. Semantic Segmentation and Diagnosis of Laryngopharyngeal Reflux Based on Class-Balanced Loss[J]. Laser & Optoelectronics Progress, 2023, 60(14): 1410020

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites