Automatic Identification of Cervical Abnormal Cells Based on Transformer

Zheng Zhang; Mingxiao Chen; Xinyu Li; Yi Chen; Shuwei Shen; Peng Yao

doi:10.3788/CJL231261

Chinese Journal of Lasers, Volume. 51, Issue 3, 0307108(2024)

Automatic Identification of Cervical Abnormal Cells Based on Transformer

Zheng Zhang¹, Mingxiao Chen¹, Xinyu Li¹, Yi Chen¹, Shuwei Shen², and Peng Yao^3、aff***

¹Department of Precision Machinery and Precision Instrumentation, School of Engineering Science, University of Science and Technology of China, Hefei 230027, Anhui , China

²Suzhou Advanced Research Institute, University of Science and Technology of China, Suzhou 215123, Jiangsu ,China

³School of Microelectronics, University of Science and Technology of China, Hefei 230027, Anhui , China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(12)

Fig. 1. Overall architecture diagram of model

Download full size

Fig. 2. Architecture diagrams of transformer encoder block. (a) Generic structure; (b) improved structure

Download full size

Fig. 3. DW convolution diagram

Download full size

Fig. 4. Changes of dynamic IOU thresholds

Download full size

Fig. 5. Comparison of detection effects of different models. (a) Ground truth; (b) proposed method; (c) Sparse R-CNN

Download full size

Fig. 6. Loss and AP under dynamic IOU threshold and fixed IOU threshold. (a) Loss; (b) AP

Download full size

Fig. 7. Ablation experiment heatmaps. (a) Original images; (b) heatmap generated by original Transformer model; (c) heatmap generated by our method

Download full size

Table 1. Experimental results of various models

View table

Table 1. Experimental results of various models

Model	AP	AP₅₀	AP₇₅	AP_S	AP_M	AP_L	Number of parameters /10⁶
YOLOv3	18.0	35.4	16.8	0.0	6.3	25.9	61.5
RetinaNet^［28］	23.5	43.7	22.5	0.3	10.6	32.2	37.9
Sparse R-CNN^［32］	24.0	42.1	23.2	0.2	10.8	33.4	108.5
Cascade R-CNN^［33］	22.1	39.3	21.4	0.2	7.6	32.6	69.3
DETR^［29］	23.5	46.3	21.6	0.5	9.2	32.6	41.5
FCOS^［30］	23.1	41.0	22.5	0.2	12.2	32.4	32.1
GiraffeDet^［31］	23.9	44.3	23.1	0.3	10.9	32.6	47.8
Iter Sparse R-CNN^［34］	24.8	43.8	25.3	0.7	11.7	33.0	123.4
Ours	26.1	46.8	25.5	1.8	13.5	33.5	48.3

Table 2. Comparison between proposed model and attFPN
View table
Table 2. Comparison between proposed model and attFPN
Model AP AP₅₀ AP₇₅
attFPN 25.0 50.3 22.2
Proposed model 26.1 46.8 25.5

Table 3. Comparison of experimental results under different backbone choices
View table
Table 3. Comparison of experimental results under different backbone choices
Backbone AP AP₅₀ AP₇₅ AP_S AP_M AP_L
CNN（Resnet-101） 23.7 42.1 24.2 1.1 12.8 31.4
Transformer 26.1 46.8 25.5 1.8 13.5 33.5

Table 4. Ablation experiment results of different modules
View table
Table 4. Ablation experiment results of different modules
Improved Transformer encoder Dynamic IOU threshold AP AP₅₀ AP₇₅ AP_S AP_M AP_L
24.1 43.8 23.8 0.5 13.2 30.7
√ 24.7 44.7 25.2 0.2 12.0 32.2
√ 25.9 46.1 25.3 2.5 12.2 33.8
√ √ 26.1 46.8 25.5 1.8 13.5 33.5

Table 5. Comparison among models based on dynamic IOU threshold and multiple fixed IOU thresholds
View table
Table 5. Comparison among models based on dynamic IOU threshold and multiple fixed IOU thresholds
IOU threshold AP AP₅₀ AP₇₅ AP_S AP_M AP_L
0.5 23.6 45.5 22.4 0.4 12.6 31.0
0.6 24.0 42.1 23.2 0.2 10.8 33.4
0.7 24.1 43.8 23.8 0.5 13.2 30.7
Dynamic threshold 24.7 44.7 25.2 0.2 12.0 32.2

Tools

Get Citation

Copy Citation Text

Zheng Zhang, Mingxiao Chen, Xinyu Li, Yi Chen, Shuwei Shen, Peng Yao. Automatic Identification of Cervical Abnormal Cells Based on Transformer[J]. Chinese Journal of Lasers, 2024, 51(3): 0307108

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites