Lung Nodule CT Image Classification Based on Adaptive Aggregate Weight Federated Learning

Fig. 10. Distribution of anti-interference experimental data. (a) Proportion of positive and negative sample data; (b) FedAaw anti-interference experiment

Download full size

Table 1. Division results of adversarial verification
View table
Table 1. Division results of adversarial verification
Center Train cohort Test cohort
LTB case LAC case LTB case LAC case
Center 1 105 155 86 156
Center 2 14 26 10 37
Center 3 32 120 25 76
Center 4 26 33 12 8
Center 5 2 37 3 46

Table 2. Comparison of AUC results of different models and Aaw threshold parameters

View table

Table 2. Comparison of AUC results of different models and Aaw threshold parameters

Center	Q=0				Q=0.5				Q=0.8
	ResNet18		MsaNet18		ResNet18		MsaNet18		ResNet18		MsaNet18
	Train	Test	Train	Test	Train	Test	Train	Test	Train	Test	Train	Test
Center 1	0.8987	0.7467	0.9393	0.7519	0.9030	0.7373	0.9459	0.7732	0.8704	0.7431	0.8938	0.8037
Center 2	0.9368	0.7622	0.9423	0.8135	0.9093	0.7919	0.9423	0.8351	0.9258	0.7784	0.9313	0.8405
Center 3	0.8341	0.6726	0.9154	0.7047	0.9276	0.7317	0.9464	0.7126	0.8638	0.7089	0.9526	0.7758
Center 4	0.8345	0.8229	0.8928	0.8542	0.9091	0.8229	0.8916	0.8542	0.8823	0.8333	0.9138	0.9063
Center 5	0.9054	0.8043	0.9459	0.8768	0.9189	0.8188	0.9189	0.8986	0.9324	0.8333	0.9459	0.9348

Table 3. When Q=0, MsaNet18 test set performance indicator

View table

Table 3. When Q=0, MsaNet18 test set performance indicator

Parameter	Center 1	Center 2	Center 3	Center 4	Center 5	Sum
Sensitivity	0.7821（122/156）	0.6486（24/37）	0.4737（36/76）	0.7500（6/8）	0.8261（38/46）	0.6996（226/323）
Specificity	0.6628（57/86）	0.9000（9/10）	0.8800（22/25）	0.8333（10/12）	1.0000（3/3）	0.7426（101/136）
Accuracy	0.7397（179/242）	0.7021（33/47）	0.5743（58/101）	0.8000（16/20）	0.8367（41/49）	0.7124（327/459）

Table 4. When Q=0.5, MsaNet18 test set performance indicator

View table

Table 4. When Q=0.5, MsaNet18 test set performance indicator

Parameter	Center 1	Center 2	Center 3	Center 4	Center 5	Sum
Sensitivity	0.8077（126/156）	0.8108（30/37）	0.6711（51/76）	0.8750（7/8）	0.8261（38/46）	0.7802（252/323）
Specificity	0.6279（54/86）	0.9000（9/10）	0.8400（21/25）	0.7500（9/12）	1.0000（3/3）	0.7059（96/136）
Accuracy	0.7438（180/242）	0.8298（39/47）	0.7129（72/101）	0.8000（16/20）	0.8367（41/49）	0.7581（348/459）

Table 5. When Q=0.8, MsaNet18 test set performance indicator

View table

Table 5. When Q=0.8, MsaNet18 test set performance indicator

Parameter	Center 1	Center 2	Center 3	Center 4	Center 5	Sum
Sensitivity	0.8333（130/156）	0.7568（28/37）	0.6579（50/76）	1.0000（8/8）	0.9348（43/46）	0.8019（259/323）
Specificity	0.6395（55/86）	0.9000（9/10）	0.8000（20/25）	0.6667（8/12）	1.0000（3/3）	0.6985（95/136）
Accuracy	0.7645（185/242）	0.7872（37/47）	0.6931（70/101）	0.8000（16/20）	0.9388（46/49）	0.7712（354/459）

Table 6. AUC results of different federal algorithms on lung nodule classification tasks

View table

Table 6. AUC results of different federal algorithms on lung nodule classification tasks

Algorithm	Center 1	Center 2	Center 3	Center 4	Center 5
FedAvg	0.6077	0.8159	0.5849	0.5303	0.6622
FedProx	0.8228	0.9890	0.8232	0.9930	0.9189
Moon	0.6371	0.8709	0.6789	0.7587	0.7027
HarmoFL	0.7799	0.9780	0.5727	0.9405	0.6216
FedAaw	0.9963	0.9988	0.9375	0.6395	0.7605

Table 7. AUC results of different federal algorithms on lung nodule classification tasks

View table

Table 7. AUC results of different federal algorithms on lung nodule classification tasks

Algorithm	Center 1	Center 2	Center 3	Center 4	Center 5
FedAvg	0.6473	0.7351	0.5932	0.4479	0.5870
FedProx	0.6506	0.7297	0.3758	0.3333	0.2899
Moon	0.6491	0.6568	0.5142	0.5652	0.7101
HarmoFL	0.6520	0.7703	0.2937	0.5313	0.3551
FedAaw	0.6943	0.7013	0.5591	0.6617	0.8478

Table 8. Results of federal anti-interference experimental training set

View table

Table 8. Results of federal anti-interference experimental training set

Distribution	Method	Center 1	Center 2	Center 3	Center 4	Center 5
Distribution 1	Distributed learning	0.8963	0.9148	0.9268	0.8858	0.8919
Distribution 1	Federated learning	0.8938	0.9313	0.9526	0.9138	0.9459
Distribution 2	Distributed learning	0.8682	0.9286	0.8770	0.8789	0.8917
Distribution 2	Federated learning	0.8816	0.9405	0.9200	0.9368	0.9417
Distribution 3	Distributed learning	0.8323	0.9306	0.9286	0.8632	0.9000
Distribution 3	Federated learning	0.9098	0.8264	0.8262	0.9053	0.9000

Table 9. Results of federal anti-interference experimental test set

View table

Table 9. Results of federal anti-interference experimental test set

Distribution	Method	Center 1	Center 2	Center 3	Center 4	Center 5
Distribution 1	Distributed learning	0.7686	0.8378	0.6821	0.7604	0.8261
Distribution 1	Federated learning	0.8037	0.8405	0.7758	0.9063	0.9348
Distribution 2	Distributed learning	0.7453	0.7405	0.6331	0.7293	0.5797
Distribution 2	Federated learning	0.7947	0.8405	0.6768	0.8672	0.9275
Distribution 3	Distributed learning	0.7624	0.7386	0.6511	0.8120	0.8478
Distribution 3	Federated learning	0.8020	0.8105	0.6875	0.8496	0.8913

Tools

Get Citation

Copy Citation Text

Jiangfeng Shi, Bao Feng, Yehang Chen, Xiangmeng Chen. Lung Nodule CT Image Classification Based on Adaptive Aggregate Weight Federated Learning[J]. Laser & Optoelectronics Progress, 2023, 60(22): 2210003

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category: Image Processing

Received: Nov. 11, 2022

Accepted: Feb. 22, 2023

Published Online: Nov. 6, 2023

The Author Email: Feng Bao (fengbao1986.love@163.com)

DOI:10.3788/LOP223027

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology

Table 1. Division results of adversarial verification

Table 1. Division results of adversarial verification

Table 2. Comparison of AUC results of different models and Aaw threshold parameters

Table 2. Comparison of AUC results of different models and Aaw threshold parameters

Table 3. When Q=0, MsaNet18 test set performance indicator

Table 3. When Q=0, MsaNet18 test set performance indicator

Table 4. When Q=0.5, MsaNet18 test set performance indicator

Table 4. When Q=0.5, MsaNet18 test set performance indicator

Table 5. When Q=0.8, MsaNet18 test set performance indicator

Table 5. When Q=0.8, MsaNet18 test set performance indicator

Table 6. AUC results of different federal algorithms on lung nodule classification tasks

Table 6. AUC results of different federal algorithms on lung nodule classification tasks

Table 7. AUC results of different federal algorithms on lung nodule classification tasks

Table 7. AUC results of different federal algorithms on lung nodule classification tasks

Table 8. Results of federal anti-interference experimental training set

Table 8. Results of federal anti-interference experimental training set

Table 9. Results of federal anti-interference experimental test set

Table 9. Results of federal anti-interference experimental test set