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

231P - CT image-based artificial intelligence quantification of intratumoral heterogeneity for predicting treatment response to immunotherapy combined with anti-angiogenic drugs in hepatocellular carcinoma

Date

07 Dec 2024

Session

Poster Display session

Presenters

Zhi-Cheng Jin

Citation

Annals of Oncology (2024) 35 (suppl_4): S1450-S1504. 10.1016/annonc/annonc1688

Authors

Z. Jin1, J. Chen2, H. Zhu3, G. Teng3

Author affiliations

  • 1 Department Of Radiology, Zhongda Hospital Affiliated to Southeast University, 210009 - Nanjing/CN
  • 2 Center Of Interventional Radiology & Vascular Surgery, Department Of Radiology, Zhongda Hospital Affiliated to Southeast University, 210009 - Nanjing/CN
  • 3 Center Of Interventional Radiology And Vascular Surgery, Department Of Radiology, Zhongda Hospital Affiliated to Southeast University, 210009 - Nanjing/CN

Resources

This content is available to ESMO members and event participants.

Abstract 231P

Background

This study aims to quantify intratumor heterogeneity and identify prognostic imaging biomarkers in patients with hepatocellular carcinoma (HCC) treated with immune checkpoint inhibitors (ICIs) combined with anti-angiogenic drugs.

Methods

This multicenter cohort study included patients with unresectable hepatocellular carcinoma who received first-line ICIs combined with anti-angiogenic drugs from January 2018 to December 2022. Primary endpoint was objective response rate. Three-dimensional tumoral regions of interest (ROIs) were delineated on pre-treatment computed tomography (CT) scans using deep learning algorithms, followed by manual calibration. A simple linear iterative clustering algorithm was used to segment super-pixel regions within ROIs. The radiomics features were extracted from ROIs and sub-regions. For feature selection and model construction, machine learning methods were applied to develop a radiomics model. Unsupervised clustering was applied to super-pixel segmentation using a Gaussian mixture model to identify similar features, and then establish a tumor habitat model. Model discrimination performance and consistency were evaluation.

Results

The study included 690 patients, with 323 patients in the training set, 139 in internal validation set, 228 in external test set. The tumor habitat model recorded AUCs of 0.98, 0.84, and 0.87 in the training, internal, and external test sets, respectively, while the tumor radiomics model achieved AUCs of 0.82, 0.86, and 0.73 in these sets. Both radiomics and habitat models outperformed clinical prediction models with good prediction consistency. Then, the high and low risk categories were identified. The tumor habitat model significantly stratified overall survival across various datasets, including 52 patients from the TCIA-LIHC cohort and 102 patients from the HCC-TACE-Seg cohort.

Conclusions

A tumor habitat model, derived from CT images, could predict the therapeutic efficacy of HCC patients treated with ICIs combined with anti-angiogenic drugs, and demonstrates superior performance compared to traditional radiomics model.

Clinical trial identification

NCT05278195.

Editorial acknowledgement

Legal entity responsible for the study

The authors.

Funding

The study was supported by National Key Research and Development Program (2018YFA0704100, 2018YFA0704104), National Natural Science Foundation of China (81827805), Jiangsu Provincial Special Program of Medical Science (BE2019750), Postgraduate Research&Practice Innovation Program of Jiangsu Province (KYCX21_0158).

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.