Related Experiment Video
Updated: Aug 6, 2026
09:03
Monitoring Functionality and Morphology of Vasculature Recruited by Factors Secreted by Fast-growing Tumor-generating Cells
Published on: November 23, 2014
Interpretable dynamic quantitative vascular morphometry features using SHAP for anti-angiogenic therapy response
1Department of Radiology, Hubei Cancer Hospital, Tongji Medical College, Huazhong University of Science and Technology, Wuhan, Hubei, China.
Science Advances
|July 24, 2026
Summary
Early detection of anti-angiogenic therapy response in lung cancer is crucial. Our machine learning model uses CT scans to predict treatment effectiveness, identifying responders with high accuracy.
Area of Science:
- Radiology and Oncology
- Artificial Intelligence in Medicine
- Medical Imaging Analysis
Background:
- Anti-angiogenic therapies show variable patient responses (40-70%), necessitating early predictive biomarkers.
- Identifying non-responders early can prevent unnecessary toxicity and guide treatment selection.
- Quantitative imaging biomarkers are needed for objective and timely assessment of treatment efficacy.
Purpose of the Study:
- To develop and validate an automated machine learning framework for early prediction of anti-angiogenic therapy response using contrast-enhanced CT.
- To identify key imaging and clinical features that differentiate responders from non-responders.
- To provide an interpretable and clinically applicable tool for personalized cancer treatment.
Main Methods:
- Development of an automated workflow integrating tumor and vessel segmentation from contrast-enhanced CT scans.
- Extraction of delta quantitative vascular morphometry features between baseline and follow-up scans.
- Building and validating machine learning models (including a delta-merge model) using fivefold cross-validation on data from 163 lung cancer patients.
- Utilizing Shapley Additive Explanations (SHAP) for model interpretability and feature attribution.
Main Results:
- The delta-merge model achieved high predictive accuracy (AUC = 0.842 internally, 0.806 externally).
- SHAP analysis revealed an "arterial-dominant, venous-adaptive" response pattern differentiating responders.
- Key features included changes in arterial involvement and venous recovery, offering imaging-visible evidence beyond traditional radiomics.
Conclusions:
- An automated machine learning framework using quantitative vascular morphometry from CT can accurately predict anti-angiogenic therapy response in lung cancer.
- The identified "arterial-dominant, venous-adaptive" pattern provides novel insights into treatment mechanisms.
- This approach supports early, imaging-based response assessment for personalized treatment strategies.
