Combination of CNN and Transformer for Lesion-Guided Honeycomb Lung CT Image Recognition

Layer	Conv1	Conv2	Conv3	Conv4
Input	224 $\times$ 224 $\times$ 3	112 $\times$ 112 $\times$ 64	64 $\times$ 64 $\times$ 128	32 $\times$ 32 $\times$ 256
Kernel	$\{\begin{matrix} 3 \times 3,64 \\ 3 \times 3,64 \end{matrix} \times 2$	$\{\begin{matrix} 3 \times 3,128 \\ 3 \times 3,128 \end{matrix} \times 2$	$\{\begin{matrix} 3 \times 3,256 \\ 3 \times 3,256 \end{matrix} \times 2$	$\{\begin{matrix} 3 \times 3,512 \\ 3 \times 3,512 \end{matrix} \times 2$
Output	112 $\times$ 112 $\times$ 64	64 $\times$ 64 $\times$ 128	32 $\times$ 32 $\times$ 256	16 $\times$ 16 $\times$ 512

Table 2. Honeycomb lung dataset partition
View table
Table 2. Honeycomb lung dataset partition
Type Training Validation Test
Normal lung 6339 2113 2113
Honeycomb lung 6063 2020 2020

Table 3. COVID-19 dataset partition
View table
Table 3. COVID-19 dataset partition
Type Training Validation Test
Normal Lung 35996 11842 12245
Pneumonia 25496 7400 7395
COVID-19 82286 6244 6018

Table 4. Classification results of deep learning algorithm on honeycomb lung dataset

View table

Table 4. Classification results of deep learning algorithm on honeycomb lung dataset

Model	$R_{a c c u r a c y}$	$R_{p r e c i s i o n}$	$R_{s e n s i t i v i t y}$	$R_{s p e c i f i c i t y}$	$s_{F 1}$
AlexNet	86.71	84.08	88.22	85.40	86.10
VGG16	90.16	90.15	89.84	90.53	89.99
ResNet101	93.44	94.16	92.56	94.31	93.35
DenseNet121	95.37	93.96	96.49	94.41	95.21
ResGNet-C	94.85	93.92	95.48	94.26	94.69
ViT	89.33	89.02	89.24	89.50	89.13
CMT-S	95.82	95.99	95.47	96.14	95.73
TransCNNNet	97.28	96.93	97.46	97.08	97.19
TransSEResNet	98.55	98.86	98.13	98.90	98.49
Proposed model	99.67	99.41	99.90	99.43	99.64

Table 5. Classification results of deep learning algorithm on COVID-CT dataset

View table

Table 5. Classification results of deep learning algorithm on COVID-CT dataset

Model	$R_{a c c u r a c y}$	$R_{p r e c i s i o n}$	$R_{s e n s i t i v i t y}$	$R_{s p e c i f i c i t y}$	$s_{F 1}$
AlexNet	94.82	95.00	87.50	98.60	91.10
VGG16	96.21	94.69	92.77	98.37	93.72
ResNet101	96.53	94.99	93.06	97.16	93.08
DenseNet121	96.57	94.53	92.84	98.40	93.68
ResGNet-C	96.62	96.21	97.33	95.91	96.65
ViT	96.60	95.30	93.80	93.20	94.60
CMT-S	97.00	97.74	97.13	99.30	97.43
TransCNNNet	96.73	97.45	97.76	96.01	96.36
TranSE-ResNet	96.76	97.11	96.61	98.55	96.85
Proposed model	97.08	98.58	97.95	98.70	98.26

Table 6. Comparison results of ablation experiments
View table
Table 6. Comparison results of ablation experiments
Baseline MHCA MSE LSG $R_{a c c u r a c y}$ /%
√ — — — 94.28
√ — √ — 97.31
√ — √ √ 98.52
√ √ √ √ 99.67

Tools

Get Citation

Copy Citation Text

Bingqian Yang, Xiufang Feng, Yunyun Dong, Yuanrong Zhang. Combination of CNN and Transformer for Lesion-Guided Honeycomb Lung CT Image Recognition[J]. Laser & Optoelectronics Progress, 2024, 61(14): 1437014

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites