Interpretability study of pretrained models via transfer learning on ImageNet for lung cancer prediction

Zhe Lin

doi:10.1117/12.2656827

20 October 2022 Interpretability study of pretrained models via transfer learning on ImageNet for lung cancer prediction

Zhe Lin

Author Affiliations +

Proceedings Volume 12451, 5th International Conference on Computer Information Science and Application Technology (CISAT 2022); 124514X (2022) https://doi.org/10.1117/12.2656827
Event: 5th International Conference on Computer Information Science and Application Technology (CISAT 2022), 2022, Chongqing, China

Abstract

Transfer learning is an effective strategy to obtain models for lung cancer analysis due to the limited number of labeled datasets. But few studies have made an interpretable explanation that why networks pretrained on ImageNet without medical images can be transferred to identify lung nodule images. In this paper, three convolutional neural networks with different depths, VGG16, MobileNet, and ResNet50, are selected for experiments to compare the performance and training difficulty between models with and without transfer learning, and to study the impact of transfer learning on networks of different depths. All the models are retrained on the same set of lung CT images with the same parameters, with and without the initial weights pretrained on the ImageNet dataset. The weights of the pretrained model are frozen, and pooling layers and classification output layers are created and concatenated to get the models required for experiments. The experimental results indicate that the initial network weights and feature extraction capabilities derived from ImageNet are applicable after being transferred to the lung nodule image set, making retraining easier and significantly improving prediction performance. In addition, transfer learning has a better effect on models with deeper networks, which contributes to alleviating model overfitting and poor generalization.

Citation Download Citation

Zhe Lin "Interpretability study of pretrained models via transfer learning on ImageNet for lung cancer prediction", Proc. SPIE 12451, 5th International Conference on Computer Information Science and Application Technology (CISAT 2022), 124514X (20 October 2022); https://doi.org/10.1117/12.2656827

ACCESS THE FULL ARTICLE

INSTITUTIONAL
Select your institution to access the SPIE Digital Library.

SELECT YOUR INSTITUTION

PERSONAL
Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.

PERSONAL SIGN IN

No SPIE Account? Create one

PURCHASE THIS CONTENT

SUBSCRIBE TO DIGITAL LIBRARY

50 downloads per 1-year subscription

Members: $195

Non-members: $335 ADD TO CART

25 downloads per 1 - year subscription

Members: $145

Non-members: $250 ADD TO CART

PURCHASE SINGLE ARTICLE

Includes PDF, HTML & Video, when available

Members: $17.00

Non-members: $21.00 ADD TO CART

PROCEEDINGS
6 PAGES

DOWNLOAD PAPER SAVE TO MY LIBRARY

GET CITATION

RIGHTS & PERMISSIONS

Get copyright permission Get copyright permission on Copyright Marketplace

KEYWORDS

Performance modeling

Lung cancer

Data modeling

Computed tomography

Medical imaging

Feature extraction

Neural networks

Show All Keywords

Keywords/Phrases

Search In:

Publication Years