Hiring in life sciences? Share your open positions with our professional community. Read more Close

Advertisement

Predicting acute radiation esophagitis after lung cancer radiotherapy: Clinical-DVH model development and incremental evaluation of deep-learning-derived dose features.

Created on 11 Nov 2026

Authors

Shaofeng Wang, Xiaolin Chen, Yangchao Xie, Xingfu Lin, Yuyin Le, Libin Liu

Published in

Medical dosimetry : official journal of the American Association of Medical Dosimetrists. Nov 10, 2026. Epub Nov 10, 2026.

Abstract

To develop and evaluate a clinical-dose-volume histogram (DVH) model for predicting grade ≥2 acute radiation esophagitis after lung cancer radiotherapy and to assess the incremental predictive information provided by deep-learning-derived dose features. This retrospective study expanded a previously reported cohort to 414 patients from Fuzhou Pulmonary Hospital of Fujian Province (Center A) and The Second Affiliated Hospital of Xiamen Medical College (Xiamen Second Hospital; Center B). Clinical predictor stability and 15 DVH metrics were assessed in 228 patients in the development cohort. The resulting clinical-DVH reference model was compared with 2 extended models incorporating either global or 4-bin longitudinal deep dose features extracted from the same 3-dimensional dose region using the same pretrained encoder with frozen weights. After selection of the clinical-DVH specification, preprocessing and ridge regression were performed within a nested cross-validation framework. The models were then evaluated without refitting in an earlier Center B cohort (n = 86) and a later Center B cohort (n = 100), which together included 110 patients not included in the previous publication. Age met the stability criterion in 91 of 100 splits. Among the clinical-DVH combinations evaluated, age plus V40 ranked highest, with a mean held-out area under the receiver operating characteristic curve (AUC) of 0.821. In the subsequent matched comparison, out-of-fold AUCs were 0.809, 0.801, and 0.787 for the reference model, global fusion, and longitudinal fusion, respectively. Corresponding AUCs in the earlier/later Center B cohorts were 0.725/0.713, 0.750/0.728, and 0.777/0.732. In the earlier/later Center B cohorts, longitudinal fusion reduced Brier scores from 0.207/0.201 to 0.174/0.178, respectively. Among the 110 additional Center B patients not included in the previous publication, its AUC point estimate was 0.635, compared with 0.576 for the reference model. Absolute risks remained overestimated, and paired bootstrap intervals for AUC differences included zero. Development-cohort analyses supported age plus V40 as a parsimonious reference model. Deep-learning-derived dose features showed potential to add predictive information at Center B, particularly with longitudinal pooling. Independent validation and further assessment of calibration are warranted.

PMID:
43009373
Bibliographic data and abstract were imported from PubMed on 11 Nov 2026.

Read full publication at:
Please sign in to see all details.

Advertisement

Stats

  • Community rating n/a 0 votes
  • Reviewers' rating n/a 0 votes
  • Your rating

1-terrible, 9-excellent. How would you rate this publication? Sign in in to submit your rating.

  • Recommendations n/a n/a positive of 0 vote(s)
  • Views 16
  • Comments 0

Recommended by

  • No recommendations yet.

Post a comment

You need to be signed in to post comments. You can sign in here.

Comments

There are no comments yet.

Advertisement