MENU

Click here to

×

Are you sure ?

Yes, do it No, cancel
  1. Home
  2. Meeting Program
  3. All
  4. Prediction of Radiation Pneumonitis After Lung Stereotactic Body Radiation Therapy Using Dosiomics Features: A Retrospective Multi-Institutional Study

Prediction of Radiation Pneumonitis After Lung Stereotactic Body Radiation Therapy Using Dosiomics Features: A Retrospective Multi-Institutional Study

T Adachi1,2*, M Nakamura1,2, T Shintani2, T Mitsuyoshi2,3, R Kakino1,2, T Ogata3, H Tanabe3, T Ono2, H Hirashima2, T Sakamoto4, M Kokubo3, Y Matsuo2, T Mizowaki2, (1) Department of Information Technology and Medical Engineering, Human Health Sciences, Graduate School of Medicine, Kyoto University, Kyoto, Japan, (2) Department of Radiation Oncology and Image-Applied Therapy, Graduate School of Medicine, Kyoto University, Kyoto, Japan, (3) Department of Radiation Oncology, Kobe City Medical Center General Hospital, Hyogo, Japan(4) Department of Radiation Oncology, Kyoto Katsura Hospital, Kyoto, Japan

Presentations

(Sunday, 7/12/2020)   [Eastern Time (GMT-4)]

Room: AAPM ePoster Library

Purpose: Radiation pneumonitis (RP) is one of the major toxicities after lung SBRT. The purpose of this study was to predict RP after lung SBRT by using dose-based radiomics features (dosiomics features).
Methods: This multi-institutional study included 247 early stage NSCLC patients who underwent SBRT with the prescribed dose of 48 to 70 Gy at the isocenter from June 2009 to March 2016. Symptomatic RP was graded according to the National Cancer Institute Common Terminology Criteria for Adverse Events v4.0 and defined RP as grade 2 or worse.
As dose-volume indices (DVIs), internal target volume (ITV) size, mean lung dose, and the whole lung excluding ITV (Lung-ITV) receiving x Gy (x=5, 10, …, 40) were used. As dosiomics features, 6,808 features (shape, first order, texture and wavelet) were extracted from inside Lung-ITV receiving x Gy.
All patients were randomly divided into the train (n=172) and the test datasets (n=75). After dosiomics features were converted to z-score, dosiomics features with Spearman’s correlation coefficients = 0.8 were removed. Afterwards, three models were built using LightGBM (Light gradient boosting machine) as follows: (?) DVI model, (?) dosiomics model and (?) hybrid model. After determining the most appropriate parameters for LightGBM by 5-fold cross validation, the final models were applied to the test datasets, and its performance was then evaluated by area under the curve (AUC) in receiver operating characteristic.
Results: Total 37 patients (15.0%) developed RP after SBRT. For the train datasets, the mean ± standard deviation of AUC with (?) DVI, (?) dosiomics and (?) hybrid model were 0.726 ± 0.028, 0.887 ± 0.076 and 0.890 ± 0.081, respectively. For the test datasets, AUC with each model were 0.673, 0.794 and 0.827, respectively.
Conclusion: We investigated dosiomics features may be more useful predictors for RP after lung SBRT than DVIs.

Keywords

Lung, Feature Extraction, Dose

Taxonomy

IM- Dataset Analysis/Biomathematics: Informatics

Contact Email

Additional Links