Reliable Polyp Segmentation Based on Local Channel Attention

Jian Xu; Ruohan Wang

doi:10.3788/LOP241160

Laser & Optoelectronics Progress, Volume. 62, Issue 2, 0217002(2025)

Reliable Polyp Segmentation Based on Local Channel Attention

Jian Xu^1,2、* and Ruohan Wang¹

Author Affiliations

¹School of Computer Science, Nanjing University of Posts and Telecommunications, Nanjing 210023, Jiangsu , China

²Jiangsu Key Laboratory of Big Data Security and Intelligent Processing (Nanjing University of Posts and Telecommunications), Nanjing 210023, Jiangsu , China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(10)

Fig. 1. Overall structure of proposed model

Download full size

Fig. 2. Schematic of LCFE

Download full size

Fig. 3. Schematic of uncertainty estimation module.

Download full size

Fig. 4. Comparison of segmentation results of different algorithms on the CVC-300 dataset

Download full size

Fig. 5. Dice results of different algorithms under different Gaussian noise levels

Download full size

Fig. 6. Comparison of segmentation results of different algorithms under different Gaussian noise levels

Download full size

Table 1. Experimental training parameter settings
View table
Table 1. Experimental training parameter settings
Setting Optimizer Learning rate Batch size Epoch Image size
‒ AdamW 1×10^-5 16 100 352×352

Table 2. Statistical data of polyp segmentation dataset
View table
Table 2. Statistical data of polyp segmentation dataset
Dataset Imaging Number of image Image size Train Test
Kvasir Colon 1000 Variable 900 100
ClinicDB Colon 612 384×288 550 62
ColonDB Colon 380 574×500 ‒ 380
CVC-300 Colon 60 574×500 ‒ 60
ETIS Colon 196 1225×966 ‒ 196

Table 3. Segmentation results of different polyp segmentation methods on five datasets

View table

Table 3. Segmentation results of different polyp segmentation methods on five datasets

Method	Kvasir		ClinicDB		CVC-300		ColonDB		ETIS
Method	Dice	mIoU	Dice	mIoU	Dice	mIoU	Dice	mIoU	Dice	mIoU
U-Net	81.80	74.60	82.30	75.50	71.00	62.70	51.20	44.40	39.80	33.50
UNet++	82.10	74.30	79.40	72.90	70.70	62.40	48.30	41.00	40.10	34.40
SFA	72.30	61.10	70.00	60.70	46.70	32.90	46.90	36.70	29.70	21.70
ParNet	89.80	84.00	89.90	84.90	87.10	79.70	71.20	64.00	62.80	56.70
SANet	90.40	84.70	91.60	85.90	88.80	81.50	75.30	67.00	75.00	65.40
Polyp-PVT	91.23	86.30	93.08	88.28	88.71	81.89	80.75	71.85	78.67	70.97
PVT-CASCADE	92.27	86.76	94.34	89.98	88.47	83.79	82.57	74.53	77.86	72.58
Proposed	92.14	86.75	93.98	89.25	89.62	84.56	82.61	74.58	81.65	73.67

Table 4. Impact of LCFE module on model performance
View table
Table 4. Impact of LCFE module on model performance
Method CVC-300 ClinicDB ETIS
Dice mIoU Dice mIoU Dice mIoU
Backbone 89.11 82.32 93.54 88.95 78.16 69.97
Backbone+LCFE 89.62 84.56 93.98 89.25 81.65 73.67