Oops, you're using an old version of your browser so some of the features on this page may not be displaying properly.

MINIMAL Requirements: Google Chrome 24+Mozilla Firefox 20+Internet Explorer 11Opera 15–18Apple Safari 7SeaMonkey 2.15-2.23

Poster Display session

370P - Transcriptome data-based prognosis prediction model for lung adenocarcinoma using an image deep learning approach

Date

28 Mar 2025

Session

Poster Display session

Presenters

hsuanyu chen

Citation

Journal of Thoracic Oncology (2025) 20 (3): S208-S232. 10.1016/S1556-0864(25)00632-X

Authors

H. chen1, Y. Yeh1, Y. Chang2

Author affiliations

  • 1 Academia Sinica, Taipei City/TW
  • 2 National Health Research Institutes - Zhunan Campus, Zhunan Township/TW

Resources

Login to get immediate access to this content.

If you do not have an ESMO account, please create one for free.

Abstract 370P

Background

In lung adenocarcinoma, therapies like EGFR TKIs and ICIs have improved outcomes for EGFR-mutant and wild-type patients, respectively. However, drug resistance and limited survival remain significant challenges. This study proposes a novel approach using whole transcriptome data to develop a prognosis prediction model through an image-based deep learning method.

Methods

Four cohorts were analyzed: one training cohort (RNA-seq, n=391) and three independent validation cohorts (TCGA RNA-seq, n=394; GSE68465, n=443; GSE13213, n=117). RNA-seq data were transformed into images using pixel encoding strategies designed to preserve transcriptomic information. The Convolutional Neural Network (CNN) model was trained on all gene expression data without feature selection. To prevent overfitting, the training cohort was divided into subsets: training (n=259), testing (n=62), and validation (n=77). The CNN model was trained on the training set, validated on the testing and validation sets, and independently evaluated on the three validation cohorts.

Results

RNA-seq data were successfully transformed into images first. The image transformation process enabled effective training of the CNN model, which categorized patients into high- and low-risk groups. In the training cohort, high-risk patients had significantly shorter overall survival (all P < 0.01). Similar results were observed in external validation cohorts: TCGA (P=0.001), GSE68465 (P < 0.0001), and GSE13213 (P=0.003). Prediction accuracies were 0.71, 0.81, and 0.92 for the training, testing, and validation subsets, respectively. In the external cohorts, prediction accuracy averaged 0.7.

Conclusions

This study introduces an innovative image-based deep learning strategy to analyze whole transcriptome data without requiring differential gene selection. This approach captures comprehensive transcriptomic information, offering potential for improved prognostic modeling and molecular guidance in lung cancer treatments.

Editorial acknowledgement

During the preparation of this work the author(s) used ChatGPT in order to do the English editing of this abstract. After using this tool/service, the author(s) reviewed and edited the content as needed and take(s) full responsibility for the content of the publication.

Funding

Academia Sinica and National Science and Technology Council (Taiwan)

Disclosure

All authors have declared no conflicts of interest.

This site uses cookies. Some of these cookies are essential, while others help us improve your experience by providing insights into how the site is being used.

For more detailed information on the cookies we use, please check our Privacy Policy.

Customise settings
  • Necessary cookies enable core functionality. The website cannot function properly without these cookies, and you can only disable them by changing your browser preferences.

  • Analytics cookies help us improve our website by collecting and reporting information on its usage.

  • We use marketing cookies to build a profile of you as a user through your actions on our site in order to better understand your interests and provide you with pertinent suggestions.