Z Song¹*, H Liu², C Li³, Y Zhang⁴, Y Jiang⁵, (1) Beihang University, Beijing, 11, CN, (2) Peking University Cancer Hospital & Institute, Beijing, 11, CN, (3) Peking University, Beijing, ,CN, (4) Key Laboratory of Carcinogenesis and Translational Research (Ministry of Education/Beijing), Peking University Cancer Hospital & Institute, Beijing, 11, CN, (5) Beihang University,

• 
  Z Song

PO-GePV-I-16 (Sunday, 7/10/2022)   [Eastern Time (GMT-4)]

ePoster Forums

Purpose: This study aimed to construct machine learning models using radiomics features in sequential cone-beam computed tomography (CBCT) for individualized and accurate estimation of radiation pneumonitis (RP) in patients with lung cancer after radiation therapy (RT).

Methods: This study retrospectively reviewed 114 patients (58 symptomatic RP patients) with lung cancer who underwent radiotherapy between 2012 and 2019. By automatically segmenting the whole lung region on the CBCT images, a total of 1288 radiomics features were extracted. Least absolute shrinkage and selection operator algorithm (Lasso) was performed to features selection and radiomics signature building. The prediction performance of multivariate logistic regression model for classifying 2 groups of patients was assessed using area under the ROC curve (AUC).

Results: The radiomics signature consisting of 19 features was significantly associated with RP status and showed great discrimination capacity with AUC 0.83 in the validation cohort.

Conclusion: The radiomics signature in this study could be used for individualized prediction of RP after RT preoperatively and help physicians adjust the treatment plan of the dose-sensitive lungs as a low-cost and non-invasive mean.

Keywords

Not Applicable / None Entered.

Taxonomy

Not Applicable / None Entered.

Contact Email