Lung Nodule Classification in CT Images Using Improved DenseNet
Objective: Computed tomography (CT) imaging of the chest is an effective diagnostic tool assisting physicians in making a diagnosis. This study aimed to propose a new convolutional neural network for classifying the lung nodules of the patient through chest CT scan data to determine whether the pati...
| Main Author: | |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Editorial Office of Advanced Ultrasound in Diagnosis and Therapy
2023-09-01
|
| Series: | Advanced Ultrasound in Diagnosis and Therapy |
| Subjects: | |
| Online Access: | https://www.journaladvancedultrasound.com/fileup/2576-2516/PDF/1696909110836-1871894835.pdf |
| _version_ | 1797639969122025472 |
|---|---|
| author | Xiuping Men, PhD, Vladimir Y. Mariano, PhD, Aihua Duan, PhD, Xiaoyan Shi, PhD |
| author_facet | Xiuping Men, PhD, Vladimir Y. Mariano, PhD, Aihua Duan, PhD, Xiaoyan Shi, PhD |
| author_sort | Xiuping Men, PhD, Vladimir Y. Mariano, PhD, Aihua Duan, PhD, Xiaoyan Shi, PhD |
| collection | DOAJ |
| description | Objective: Computed tomography (CT) imaging of the chest is an effective diagnostic tool assisting physicians in making a diagnosis. This study aimed to propose a new convolutional neural network for classifying the lung nodules of the patient through chest CT scan data to determine whether the patient has related disease genes. Methods: We proposed a DenseNet-based neural network structure that uses multi-scale convolutional kernels to obtain features of different receptive fields, which are fed into a DenseNet containing four improved DenseBlocks, followed by a classification module to obtain the model output, i.e., whether a lung nodule contains a cancer gene. We conducted classification experiments on a CT scan dataset containing 465 training samples and 117 test samples. Results: The results showed that DenseNet was better than ResNet in terms of classification, whereas ResNet was better than VGG, which was consistent with the findings of previous studies. However, because these models were more complex, they suffered from overfitting problems. Among all of the models used in this paper, our proposed network achieved the best results in terms of accuracy, F1 score, and sensitivity without an over fitting. The accuracy was 72.0%, sensitivity was 78%, and F1 score was 68%. Conclusion: The proposed DenseNet neural network can improve and assist medical imaging diagnostic physicians in the initial diagnosis of lung nodules. |
| first_indexed | 2024-03-11T13:24:11Z |
| format | Article |
| id | doaj.art-1754c913d17c4406bc089046626bf12b |
| institution | Directory Open Access Journal |
| issn | 2576-2516 |
| language | English |
| last_indexed | 2024-03-11T13:24:11Z |
| publishDate | 2023-09-01 |
| publisher | Editorial Office of Advanced Ultrasound in Diagnosis and Therapy |
| record_format | Article |
| series | Advanced Ultrasound in Diagnosis and Therapy |
| spelling | doaj.art-1754c913d17c4406bc089046626bf12b2023-11-03T08:33:51ZengEditorial Office of Advanced Ultrasound in Diagnosis and TherapyAdvanced Ultrasound in Diagnosis and Therapy2576-25162023-09-017327227810.37015/AUDT.2022.220018Lung Nodule Classification in CT Images Using Improved DenseNetXiuping Men, PhD, Vladimir Y. Mariano, PhD, Aihua Duan, PhD, Xiaoyan Shi, PhD0a School of Management Science and Engineering, Anhui University of Finance and Economics, Bengbu, Anhui, China;b School of Computing and Information Technologies National University, Manila, PhilippinesObjective: Computed tomography (CT) imaging of the chest is an effective diagnostic tool assisting physicians in making a diagnosis. This study aimed to propose a new convolutional neural network for classifying the lung nodules of the patient through chest CT scan data to determine whether the patient has related disease genes. Methods: We proposed a DenseNet-based neural network structure that uses multi-scale convolutional kernels to obtain features of different receptive fields, which are fed into a DenseNet containing four improved DenseBlocks, followed by a classification module to obtain the model output, i.e., whether a lung nodule contains a cancer gene. We conducted classification experiments on a CT scan dataset containing 465 training samples and 117 test samples. Results: The results showed that DenseNet was better than ResNet in terms of classification, whereas ResNet was better than VGG, which was consistent with the findings of previous studies. However, because these models were more complex, they suffered from overfitting problems. Among all of the models used in this paper, our proposed network achieved the best results in terms of accuracy, F1 score, and sensitivity without an over fitting. The accuracy was 72.0%, sensitivity was 78%, and F1 score was 68%. Conclusion: The proposed DenseNet neural network can improve and assist medical imaging diagnostic physicians in the initial diagnosis of lung nodules.https://www.journaladvancedultrasound.com/fileup/2576-2516/PDF/1696909110836-1871894835.pdf|computed tomography (ct) imaging|lung nodule classification |
| spellingShingle | Xiuping Men, PhD, Vladimir Y. Mariano, PhD, Aihua Duan, PhD, Xiaoyan Shi, PhD Lung Nodule Classification in CT Images Using Improved DenseNet Advanced Ultrasound in Diagnosis and Therapy |computed tomography (ct) imaging|lung nodule classification |
| title | Lung Nodule Classification in CT Images Using Improved DenseNet |
| title_full | Lung Nodule Classification in CT Images Using Improved DenseNet |
| title_fullStr | Lung Nodule Classification in CT Images Using Improved DenseNet |
| title_full_unstemmed | Lung Nodule Classification in CT Images Using Improved DenseNet |
| title_short | Lung Nodule Classification in CT Images Using Improved DenseNet |
| title_sort | lung nodule classification in ct images using improved densenet |
| topic | |computed tomography (ct) imaging|lung nodule classification |
| url | https://www.journaladvancedultrasound.com/fileup/2576-2516/PDF/1696909110836-1871894835.pdf |
| work_keys_str_mv | AT xiupingmenphdvladimirymarianophdaihuaduanphdxiaoyanshiphd lungnoduleclassificationinctimagesusingimproveddensenet |