Application and limitation of radiomics approach to prognostic prediction for lung stereotactic body radiotherapy using breath‐hold CT images with random survival forest: A multi‐institutional study

Abstract

Purpose: To predict local recurrence (LR) and distant metastasis (DM) in early stage non-small cell lung cancer (NSCLC) patients after stereotactic body radiotherapy (SBRT) in multiple institutions using breath-hold computed tomography (CT)-based radiomic features with random survival forest. Methods: A total of 573 primary early stage NSCLC patients who underwent SBRT between January 2006 and March 2016 and met the eligibility criteria were included in this study. Patients were divided into two datasets: training (464 patients in 10 institutions) and test (109 patients in one institution) datasets. A total of 944 radiomic features were extracted from manually segmented gross tumor volumes (GTVs). Feature selection was performed by analyzing inter-segmentation reproducibility, GTV correlation, and inter-feature redundancy. Nine clinical factors, including histology and GTV size, were also used. Three prognostic models (clinical, radiomic, and combined) for LR and DM were constructed using random survival forest (RSF) to deal with total death as a competing risk in the training dataset. Robust models with optimal hyper-parameters were determined using fivefold cross-validation. The patients were dichotomized into two groups based on the median value of the patient-specific risk scores (high-and low-risk score groups). Gray's test was used to evaluate the statistical significance between the two risk score groups. The prognostic power was evaluated by the concordance index with the 95% confidence intervals (CI) via bootstrapping (2000 iterations). Results: The concordance indices at 3 yr of clinical, radiomic, and combined models for LR were 0.57 [

Multi‐institutional dose‐segmented dosiomic analysis for predicting radiation pneumonitis after lung stereotactic body radiation therapy

“…The z‐scores are the standardized values with a mean of zero and a standard deviation (SD) of one to put different variables on the same scale. If a high correlation exists between two features (CC ≥ 0.8), then one of the two features that are highly correlated with the other remaining features is eliminated 18,26,27 . The threshold value above was determined to eliminate the multicollinearity of the features for the next step.…”

“…If a high correlation exists between two features (CC ≥ 0.8), then one of the two features that are highly correlated with the other remaining features is eliminated. 18,26,27 The threshold value above was determined to eliminate the multicollinearity of the features for the next step. Second, important features were selected using the "Boruta" package based on the LightGBM, which is widely used for binary classifications.…”

“…14 In particular, a radiomic approach that enables the extraction of quantitative medical imaging features for predicting prognostic outcomes was applied to computed tomography (CT) images to predict patient prognosis, such as local recurrence and distant metastasis after lung SBRT. [15][16][17][18] Similar to the CT-based radiomic approach, some studies have reported the effectiveness of a dose-based radiomic (dosiomic) approach based on three-dimensional (3D) dose distributions to predict toxicities after radiation therapy. [19][20][21][22] In addition to the limited DVIs, dosiomic features provide quantitative analysis based on multidimensional data, such as the shape, statistics, and texture features of the dose distribution.…”

“…Recently, machine learning has progressed rapidly in the field of radiation oncology 14 . In particular, a radiomic approach that enables the extraction of quantitative medical imaging features for predicting prognostic outcomes was applied to computed tomography (CT) images to predict patient prognosis, such as local recurrence and distant metastasis after lung SBRT 15–18 . Similar to the CT‐based radiomic approach, some studies have reported the effectiveness of a dose‐based radiomic (dosiomic) approach based on three‐dimensional (3D) dose distributions to predict toxicities after radiation therapy 19–22 .…”

Impact of feature selection methods and subgroup factors on prognostic analysis with CT-based radiomics in non-small cell lung cancer patients

“…As long as clinical predictors significantly affect prognosis [17,21,35], the most representative clinical predictors were added to the features used in this study. We used a total of eight clinical predictors, namely, gender, age, each TNM stage, clinical stage, histological subtype, and chemotherapy status [36][37][38].…”