Optics and Precision Engineering, Volume. 31, Issue 7, 1074(2023)
Pneumonia aided diagnosis model based on dense dual-stream focused network
Figures & Tables(13)
Fig. 7. ROC curves for each model in small target pneumonia X-ray dataset
Table 1. Comparison of results of ablation experiments
View tableView in ArticleTable 1. Comparison of results of ablation experiments
模 型 参数量 （ M）计算量 （ G）训练
时间
/ s准确率/% 精确率/% 召回率/% F1分数/% AUC值/% DenseNet121［ 4 ］60.88 5.732 16 670 93.44±2.1 96.04±1.8 92.69±1.6 94.33±1.7 93.60±2.0 实验一 58.25 5.563 16 591 94.63±1.5 96.46±1.2 94.60±1.1 95.41±1.2 94.64±1.3 实验二 81.19 4.382 15 934 96.35±1.9 97.07±1.7 96.74±1.6 96.90±1.7 96.27±2.0 实验二 76.31 4.068 15 672 96.89±1.8 97.41±1.5 97.30±1.4 97.36±1.5 96.79±1.8 实验三 76.31 4.149 15 729 98.01±1.1 98.53±0.8 98.09±0.9 98.31±0.8 97.99±1.0
Table 2. Specific results of each model in pneumonia X-ray dataset
View tableView in ArticleTable 2. Specific results of each model in pneumonia X-ray dataset
模 型 参数量（M） 计算量（G） 训练
时间/s
准确率/% 精确率/% 召回率/% F1分数/% AUC值/% VGG19［ 5 ］1 096.09 39.294 26 613 89.07±2.4 94.81±2.2 86.16±2.9 90.28±2.2 89.69±2.8 ResNet101［ 8 ］324.28 15.664 21 291 91.32±1.5 93.46±1.5 91.68±1.7 92.56±1.3 91.24±1.9 DenseNet121［ 4 ］60.88 5.732 16 670 93.44±2.1 96.04±1.8 92.69±1.6 94.33±1.7 93.60±2.0 SeResNet101［ 8 ］360.47 15.692 22 413 93.83±1.3 95.12±1.2 94.38±1.5 94.75±1.3 93.72±1.4 EfficientNetb0［ 18 ］30.59 0.027 15 155 91.52±2.5 94.61±2.4 90.78±3.1 92.65±2.6 91.68±2.8 EfficientNetb4［ 11 ］133.91 0.068 19 299 93.97±1.9 96.18±1.7 93.48±1.5 94.81±1.7 94.08±1.9 RegNetx032［ 18 ］72.66 3.955 18 492 94.63±1.6 96.01±1.5 94.83±1.6 95.42±1.8 94.59±1.7 SwinTransformer［ 14 ］411.75 18.836 24 327 95.29±2.1 95.75±2.0 96.29±2.2 96.02±2.1 95.08±2.3 NATransformer［ 19 ］240.63 9.308 19 529 95.63±1.8 96.71±1.6 95.84±1.9 96.27±1.7 95.58±1.9 DDSF-Net 76.31 4.149 15 729 98.01±1.1 98.53±0.8 98.09±0.9 98.31±0.8 97.99±1.0
Table 3. Specifications of different models in small target pneumonia X-ray dataset
View tableView in ArticleTable 3. Specifications of different models in small target pneumonia X-ray dataset
模 型 准确率 精确率 召回率 F1 AUC值 ResNet101［ 8 ］86.24 90.00 88.73 89.36 85.16 DenseNet121［ 4 ］88.07 92.03 89.44 90.72 87.48 RegNetx032［ 18 ］90.83 94.20 91.55 92.86 90.51 NATransformer［ 19 ］92.66 93.75 95.07 94.41 91.61 DDSF-Net 95.41 97.14 95.77 96.45 95.26
Table 4. Deep learning model for diagnosing COVID-19 X-ray images
View tableView in ArticleTable 4. Deep learning model for diagnosing COVID-19 X-ray images
模 型 敏感度 特异度 准确率 精确率 F1 VGG19［ 5 ］82.96 93.96 92.33 - - Covid-caps［ 3 ］90.00 95.80 95.70 - - DCNN［ 6 ］97.91 91.87 - 93.36 - GSEN［ 20 ］93.40 98.08 95.60 92.73 95.50 DDSF-Net 98.36 98.08 98.23 98.08 98.08
Table 5. Deep learning model for diagnosing pneumonia X-ray images
View tableView in ArticleTable 5. Deep learning model for diagnosing pneumonia X-ray images
模 型 敏感度 特异度 准确率 AUC DenseNet121［ 4 ］91.00 87.00 88.13 90.00 EfficientNetb5［ 11 ］83.00 92.00 94.64 95.00 Covid-caps［ 3 ］90.00 95.00 95.00 97.00 ViT-B32［ 21 ］96.00 96.00 96.00 99.10 DDSF-Net 98.55 98.97 98.71 99.53
Table 6. Deep learning model for diagnosing pneumonia X-ray images
View tableView in ArticleTable 6. Deep learning model for diagnosing pneumonia X-ray images
模 型 准确率 AUC Dark COVID-Net［ 13 ］0.870 2 - AF-CAP［ 22 ］0.991 6 0.988 0 DDSF-Net 0.996 3 0.991 7
Tools
Get Citation
Copy Citation Text
Tao ZHOU, Xinyu YE, Huiling LU, Yuncan LIU, Xiaoyu CHANG. Pneumonia aided diagnosis model based on dense dual-stream focused network[J]. Optics and Precision Engineering, 2023, 31(7): 1074
Paper Information
Category: Information Sciences
Received: Sep. 28, 2020
Accepted: --
Published Online: Apr. 28, 2023
The Author Email: Xinyu YE (3303626778@qq.com)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20